Revert "[MLIR][TORCH] Only unroll prim loop-like ops within a torch.shape.calculate region"

[AmosLewis]

Issue found when bump torch-mlir in iree iree-org/iree#18992. This revert can fix the issue iree-org/iree#18992 (comment) when bump.
This reverts commit 55ff110.
@Max191 has done this in last bump https://github.com/iree-org/torch-mlir/commits/integrates/llvm-20241024/ , but just in iree-org/torch-mlir, the revert patch has not been added in llvm/torch-mlir.

[zjgarvey]

What is the issue?

[jinchen62]

LGTM. Might need @zjgarvey to take a look.

[AmosLewis]

What is the issue?

@zjgarvey Slack has more info.

model.mlir:4:10: error: failed to legalize operation 'arith.sitofp' that was explicitly marked illegal
    %0 = torch.operator "onnx.TfIdfVectorizer"(%arg0) {torch.onnx.max_gram_length = 2 : si64, torch.onnx.max_skip_count = 0 : si64, torch.onnx.min_gram_length = 2 : si64, torch.onnx.mode = "TF", torch.onnx.ngram_counts = [0 : si64, 4 : si64], torch.onnx.ngram_indexes = [0 : si64, 1 : si64, 2 : si64, 3 : si64, 4 : si64, 5 : si64, 6 : si64], torch.onnx.pool_int64s = [2 : si64, 3 : si64, 5 : si64, 4 : si64, 5 : si64, 6 : si64, 7 : si64, 8 : si64, 6 : si64, 7 : si64]} : (!torch.vtensor<[2,6],si32>) -> !torch.vtensor<[2,7],f32> 
         ^
model.mlir:4:10: note: see current operation: %327 = "arith.sitofp"(<<UNKNOWN SSA VALUE>>) : (i64) -> f64
model.mlir:1:1: error: conversion to vm.module failed

[zjgarvey]

Yeah let me take a closer look. It doesn't make sense to revert this patch.

[zjgarvey]

I'm going to say we should not be reverting this.

If IREE is unable to manage something about scalar scf.for loops, then we should make a change in IREE to address this, rather than going back to indiscriminately unrolling IR loops during a pass that is supposed to perform shape inference in torch-mlir.

Alternatively, if something is wrong with TFIDF, we should be resolving that instead of reverting this patch.

Max posted an issue in IREE to track iree-org/iree#18961. I suggest we mark those three tests batched TFIDF tests as failures (regressions) in IREE. Then we can actually get someone to resolve it. Or not, since I've yet to see TFIDF show up in a single model.

@ScottTodd @Max191 Do you guys have any strong opinions on this?