[chore]: Make TSConfig stricter and fix as many errors as possible

[inlined]

As part of #1591 I've added a bit more stringency to Genkit's tsconfigs. This is required to add a genkit dependency to firebase-functions as it currently compiles cleanly with skipLibCheck: false explicitly in customers' default templates; asking customers to change this would be a breaking change.

Common fixes included removing unused or non-existent imports/exports, removing dead code, typing the results of response.json(), adding missing libraries, and in one case adding a missing return statement. I started by modifying the base tsconfig.json and then reenabling (only) skipLibCheck on packages that are broken due to downstream dependencies, with a comment describing the reason why the setting is necessary. As a follow-up, we can file bugs and/or make contributions to those downstream libraries to further lessen our dependency on skipLibCheck.

This change ensures skipLibCheck is not necessary for the following packages:

• genkit
• @genkit-ai/ai
• @genkit-ai/core
• @genkit-ai/checks
• @genkit-ai/dev-local-vectorstore
• @genkit-ai/dotprompt
• @genkit-ai/evaluators
• @genkit-ai/express
• @genkit-ai/googleai
• @genkit-ai/ollama
• @genkit-ai/firebase
• @genkit-ai/google-cloud

This change does not fix the following pacakges:

• @genkit-ai/chroma
• @genkit-ai/langchain
• @genkit-ai/pinecone
• @genkit-ai/vertexai
• @genkit-ai/mcp

Issues with the remaining packages:

• firebase, google-cloud, langchain are all blocked on the fact that "long" (a dependency of protobufjs) doesn't do import/require correctly EDIT firebase and google-cloud were fixed with an update to the long library, but langchain has a smattering of errors listed below.
• chromadb triest to import packages that are not shipped with it (I suspect they need to move their types package from a devDep to a straight dep)
• pinecone tries to import types that don't exist
• MCP seems to be blocked on us. We have one error that assertNotificationCompatibility is non-conforming because our method type is never. We have four errors because ts-up is turning our type imports into normal imports and dropping the with { 'resolution-mode': 'import' } annotation, which breaks the CJS .d.ts. I've confirmed that editing the output lib file fixes the issue but it doesn't seem ts-up can be made to preserve imports correctly.

Checklist (if applicable):

• PR title is following https://www.conventionalcommits.org/en/v1.0.0/
• Tested (Test app below compiles with skipLibCheck false when installing from packed tarballs)
• Docs updated (N/A)

Tested by verifying the following program compiles as both a commonJS and ESM module:

import { Document, genkit, z, ModelArgument } from "genkit";
import { BaseEvalDataPoint, Score } from "genkit/evaluator";
import { googleAI } from "@genkit-ai/googleai";
import { ollama } from "genkitx-ollama";
import { checksMiddleware, ChecksEvaluationMetricType } from "@genkit-ai/checks";
import { devLocalIndexerRef } from "@genkit-ai/dev-local-vectorstore";

const ai = genkit({
  promptDir: ".",
  plugins: [
   googleAI(),
   ollama({
      models: [
         {
            name: "gemma",
            type: "generate",
         }
      ],
      serverAddress: "http://127.0.0.1:11434",
   }),
  ],
});

const indexer = devLocalIndexerRef("Dadabase");

(async () => {
   await ai.index({
      indexer,
      documents: [
         Document.fromText("What is Beethoven's favorite fruit? Bananana!"),
         Document.fromText("What do lawyers wear to court? Lawsuits!"),
         Document.fromText("How do astronauts pay for coffee? Star bucks!"),
      ],
   });
});

export const jokeTeller = ai.defineFlow({
   name: "jokeTeller",
   inputSchema: z.string().optional(),
   outputSchema: z.string(),
}, async (type, cb) => {
   const { stream, response } = await ai.generateStream({
     prompt: `Tell me a ${type ?? "dad"} joke.`,
     use: [
      checksMiddleware({
         authOptions: { projectId: "project-id" },
         metrics: [
            ChecksEvaluationMetricType.SEXUALLY_EXPLICIT,
            ChecksEvaluationMetricType.HATE_SPEECH,
         ],
      }),
     ],
   });
   for await (const chunk of stream) {
      cb(chunk);
   }
   return (await response).text;
})

const isFunny = ai.prompt("isFunny");

export async function funnyScore<CustomModelOptions extends z.ZodTypeAny>(
   judgeLlm: ModelArgument<CustomModelOptions>,
   dataPoint: BaseEvalDataPoint,
   judgeConfig?: CustomModelOptions
): Promise<Score> {
   const response = await ai.generate(isFunny.render({output: dataPoint.output!}));
   return {
      score: response.output,
      details: { reasoning: "because this is fake" },
   };
}

[inlined]

Doing a social network crawl, I've managed to get ahold of someone who could fix the long library. This should fix the firebase, google-cloud, and langchain plugins. Will extend my tests to verify.

[inlined]

Welp; langchain produced a lot of errors still:

node_modules/@langchain/core/dist/language_models/chat_models.d.ts:75:129 - error TS2304: Cannot find name 'this'.

75     _generateCached({ messages, cache, llmStringKey, parsedOptions, handledOptions, }: ChatModelGenerateCachedParameters<typeof this>): Promise<LLMResult & {
                                                                                                                                   ~~~~

node_modules/@langchain/core/dist/language_models/llms.d.ts:67:129 - error TS2304: Cannot find name 'this'.

67     _generateCached({ prompts, cache, llmStringKey, parsedOptions, handledOptions, runId, }: LLMGenerateCachedParameters<typeof this>): Promise<LLMResult & {
                                                                                                                                   ~~~~

node_modules/@langchain/core/tracers/base.d.cts:1:15 - error TS1479: The current file is a CommonJS module whose imports will produce 'require' calls; however, the referenced file is an ECMAScript module and cannot be imported with 'require'. Consider writing a dynamic 'import("../dist/tracers/base.js")' call instead.

1 export * from '../dist/tracers/base.js'
                ~~~~~~~~~~~~~~~~~~~~~~~~~

node_modules/@langchain/openai/dist/azure/chat_models.d.ts:2:10 - error TS2305: Module '"@langchain/core/language_models/chat_models"' has no exported member 'LangSmithParams'.

2 import { LangSmithParams, type BaseChatModelParams } from "@langchain/core/language_models/chat_models";
           ~~~~~~~~~~~~~~~

node_modules/@langchain/openai/dist/chat_models.d.ts:6:25 - error TS2305: Module '"@langchain/core/language_models/chat_models"' has no exported member 'LangSmithParams'.

6 import { BaseChatModel, LangSmithParams, type BaseChatModelParams } from "@langchain/core/language_models/chat_models";
                          ~~~~~~~~~~~~~~~

node_modules/langchain/evaluation.d.cts:1:15 - error TS1479: The current file is a CommonJS module whose imports will produce 'require' calls; however, the referenced file is an ECMAScript module and cannot be imported with 'require'. Consider writing a dynamic 'import("./dist/evaluation/index.js")' call instead.

1 export * from './dist/evaluation/index.js'
                ~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I got firebase and google-cloud to work, though I had to kick some lockfiles. It might be worth someone else double checking my work. New code added:

import { enableFirebaseTelemetry } from "@genkit-ai/firebase";
import { enableGoogleCloudTelemetry } from "@genkit-ai/google-cloud";

enableFirebaseTelemetry();
enableGoogleCloudTelemetry();