SNOW-1445416, SNOW-1445419: Implement DataFrame/Series.attrs

[sfc-gh-joshi]

1. Which Jira issue is this PR addressing? Make sure that there is an accompanying issue to your PR.

   Fixes SNOW-1445416 and SNOW-1445419

2. Fill out the following pre-review checklist:

   • I am adding a new automated test(s) to verify correctness of my new code
     • If this test skips Local Testing mode, I'm requesting review from @snowflakedb/local-testing
   • I am adding new logging messages
   • I am adding a new telemetry message
   • I am adding new credentials
   • I am adding a new dependency
   • If this is a new feature/behavior, I'm adding the Local Testing parity changes.
   • I acknowledge that I have ensured my changes to be thread-safe

3. Please describe how your code solves the related issue.

Implements DataFrame/Series.attrs by adding a new query compiler variable _attrs that is read out by frontend objects. A new annotation on the query compiler, _propagate_attrs_on_methods, will either copy _attrs from self to the return value, or reset _attrs on the return value.

I initially intended to implement this solely at the frontend layer with the override system (similar to how telemetry is added to all methods), but this created difficulties when preserving attrs across in-place operations like df.columns = [...], and could create ambiguity if the frame had a column named "_attrs". Implementing propagation at the query compiler level is simpler.

This PR also adds a new test_attrs=True parameter to eval_snowpark_pandas_result. eval_snowpark_pandas_result will set a dummy value of attrs on its inputs, and ensure that if the result is a DF/Series, the attrs field on the result matches that of pandas. Since pandas isn't always consistent about whether it propagates attrs or resets it (for some methods, the behavior depends on the input, and for some methods, it is inconsistent between Series/DF), setting test_attrs=False skips this check. When I encountered such inconsistent methods, I elected to have Snowpark pandas always propagate attrs, since it seems unlikely that users would rely on the attrs of a result being empty if they did not explicitly set it.

[sfc-gh-helmeleegy]

LGTM. Great work adding all the tests!

[sfc-gh-nkumar]

What is this check len(self.attrs) for?

[sfc-gh-joshi]

It checks that self._attrs was not empty, and avoids setting result._attrs if so, since intermediate calls to query compiler operations may have already set result._attrs = self._attrs. This lets us potentially avoid creating a new empty dict for result._attrs if it's already empty (not a very important optimization, but I think it's good practice anyway).