Large number of iids presenting new challenges

[jbusecke]

#145 seems to have unblocked a lot of the iids that previously were not available! Big Win in general, but we need to work on some of the parts of our infrastructure.

This has led to two issues:

• The BQ client can not handle more than 10k iids to check the database against failed here. Solving Implement automatic batching for CMIPBQInterface.iid_list_exists leap-data-management-utils#33 should resolve this
• The new API client hit a wall with too many dataset_ids passed. I have implemented a batching logic in the PR branch New Async API client jbusecke/pangeo-forge-esgf#45. This is already suuuper slow, and it might be worth investing a day to refactor this logic to async.

The speed considerations will become more pertinent as we add more iids with time. In particular the 'parsing' step where we go from the input list (with wildcards, brackets) to a list of single iids will produce more and more requests on each run of the deployment action.
The following steps will presumably get more manageable over time since we are pruning off the iids that are already ingested.

We are currently also handling this fairly inefficiently and are basically querying for the dataset info twice (once in expand_instance_id_list and then in get_recipe_inputs_from_iid_list(which currently takes a list of instance ids).

Going forward we should probably extract something like
{
'instance_id': {'id':..., 'field_a':..., },
'other_instance_id':{'id':..., 'field_a':..., },
...
}

This would make it trivial to prune off existing iids and then passing only the 'id' fields to get_recipe_inputs_from_iid_list

[jbusecke]

#149 does implement the bq batching in the recipe, but still waiting for a proper fix in leap-stc/leap-data-management-utils#33

[jbusecke]

Waiting on #149 merged results to see if we can handle a bunch more iids (even though it might be slow for now).

[jbusecke]

https://github.com/leap-stc/cmip6-leap-feedstock/actions/runs/8990218806 is actually highly successful! 200+ datasets ingested and still going strong!