fix(pandas): Support pandas v3 timestamp semantics while preserving v2 behavior (DM-3534)

[haakonvt]

Enable pandas v3 compatibility by making SDK timestamp resolution version-conditional ("ns" on v2, "ms" on v3), remove the pandas upper bound, update migration/docs and tests, and cache version/unit helper lookups.

[gemini-code-assist]

Code Review

This pull request updates the SDK to support pandas v3 by dynamically adjusting timestamp precision (nanoseconds for pandas v2 and lower, and milliseconds for pandas v3+). Feedback on the changes highlights a bug in _pandas_helpers.py where np.ndarray timestamps are parsed without specifying unit="ms", and suggests using the PANDAS_TS_UNIT constant in integration tests to avoid hardcoding the timestamp unit.

[audunska]

Haven't really reviewed anything yet, but shouldn't we have a strict upper bound still? Like, >= 2.1, <4

[haakonvt]

No, but it sure is up for discussion - and has been in the past.

Both numpy and pandas are very mature and responsible packages that we used to have open upper bounds for. It was only when v3 of pandas came out with what I deemed "a potentially breaking change for us" that we temporarily forced major version == 2 until it was addressed by us (aka as of this PR).

I didn't find the original discussion, but I found this; best summarised as "lock out future major releases just in case":
https://cognitedata.slack.com/archives/CG10VQPFX/p1773826110266419?thread_ts=1773760929.634279&cid=CG10VQPFX

[codecov]

Codecov Report

❌ Patch coverage is 99.15254% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 93.04%. Comparing base (a72fee9) to head (c88a5cf).

Files with missing lines	Patch %	Lines
tests/tests_unit/test_utils/test_pandas_helpers.py	98.59%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2780   +/-   ##
=======================================
  Coverage   93.04%   93.04%           
=======================================
  Files         515      516    +1     
  Lines       53006    53097   +91     
=======================================
+ Hits        49317    49404   +87     
- Misses       3689     3693    +4     

Files with missing lines	Coverage Δ
cognite/client/_api/datapoints.py	97.66% <100.00%> (+<0.01%)	⬆️
cognite/client/_api/raw/rows.py	91.07% <100.00%> (+0.05%)	⬆️
cognite/client/_sync_api/datapoints.py	100.00% <100.00%> (ø)
cognite/client/_sync_api/raw/rows.py	100.00% <ø> (ø)
cognite/client/data_classes/datapoints.py	90.13% <100.00%> (ø)
cognite/client/utils/_pandas_helpers.py	92.18% <100.00%> (+1.12%)	⬆️
...tegration/test_api/test_datapoint_subscriptions.py	98.45% <100.00%> (ø)
...ests/tests_integration/test_api/test_datapoints.py	99.31% <100.00%> (+<0.01%)	⬆️
tests/tests_unit/test_api/test_datapoints.py	99.42% <100.00%> (ø)
tests/tests_unit/test_api/test_raw.py	100.00% <ø> (ø)
... and 7 more

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[haakonvt]

/gemini review

[gemini-code-assist]

Code Review

This pull request updates the SDK to support Pandas v3 by dynamically adjusting the precision of generated timestamp columns and indices (nanoseconds for Pandas v2 and milliseconds for Pandas v3) and updates dependency constraints to allow pandas >=2.1. Feedback on the changes suggests broadening the type hint of to_pandas_timestamp to int | float | None to prevent type-checking errors with optional timestamps, and adding a pytest.importorskip guard to the new test file to avoid failures during core-only test runs.