fix(pandas): version-conditional timestamp resolution for to_pandas() (ns on v2, ms on v3)

[haakonvt]

Pandas v3 removed the implicit nanosecond default for datetimes, so to_pandas() output would get arbitrary precision depending on input. This pins resolution explicitly: ns on pandas v2 (matching old behavior), ms on pandas v3 (matching CDF's native resolution), via shared helpers in _pandas_helpers.py used across every timestamp-producing code path.

"Single" resource objects (aka not the list version), now delegate to the list-class to_pandas method to avoid having duplicated logic. This relies on _LIST_CLASS / _RESOURCE back-references staying in sync, so test_meta.py now enforces it: every CogniteResourceList._RESOURCE must point to a resource class whose _LIST_CLASS points right back. A class added without wiring up its counterpart now fails CI instead of silently duplicating conversion logic later. Also adds a test_standalone_to_pandas_allowlist test documenting the deliberate exceptions (e.g. Datapoint, RowCore) that don't follow this pattern.

For the pandas versioning itself, added tests/tests_unit/test_utils/test_pandas_helpers.py with a force_pandas_major_version fixture that monkeypatches the version check, so both the ns and ms branches run in every CI invocation regardless of which pandas is actually installed.

[gemini-code-assist]

Code Review

This pull request standardizes the to_pandas export behavior across the SDK to use millisecond precision (datetime64[ms]) for timestamp and datetime columns, aligning with the Cognite API and ensuring compatibility with pandas v3. To eliminate duplicate conversion logic, CogniteResource.to_pandas now delegates to its corresponding list class via a new _LIST_CLASS class variable, with a fallback helper for resource types without a list counterpart. Custom to_pandas overrides (such as in Group and Instances) have been simplified or removed in favor of this unified delegation pattern, and meta-tests have been added to enforce the back-references between resource and list classes. No review comments were provided, so there is no feedback to assess.

[codecov]

Codecov Report

❌ Patch coverage is 99.16318% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.69%. Comparing base (8b99b98) to head (cf6d97b).

Files with missing lines	Patch %	Lines
cognite/client/utils/_pandas_helpers.py	90.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2724      +/-   ##
==========================================
- Coverage   93.71%   93.69%   -0.03%     
==========================================
  Files         504      504              
  Lines       51294    51476     +182     
==========================================
+ Hits        48070    48229     +159     
- Misses       3224     3247      +23     

Files with missing lines	Coverage Δ
cognite/client/data_classes/_base.py	93.70% <100.00%> (-0.07%)	⬇️
cognite/client/data_classes/agents/agent_tools.py	98.67% <100.00%> (+0.01%)	⬆️
cognite/client/data_classes/agents/agents.py	100.00% <100.00%> (ø)
cognite/client/data_classes/agents/chat.py	97.11% <100.00%> (+0.01%)	⬆️
cognite/client/data_classes/aggregations.py	89.32% <100.00%> (+0.04%)	⬆️
cognite/client/data_classes/annotations.py	97.89% <100.00%> (+0.04%)	⬆️
cognite/client/data_classes/assets.py	94.47% <100.00%> (+0.02%)	⬆️
cognite/client/data_classes/capabilities.py	97.01% <100.00%> (+<0.01%)	⬆️
cognite/client/data_classes/contextualization.py	91.24% <100.00%> (+0.05%)	⬆️
...te/client/data_classes/data_modeling/containers.py	97.70% <100.00%> (+0.01%)	⬆️
... and 59 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.