Fix _drop_timestamp being called as an instance method

[magic-peach]

What does this PR do?

_drop_timestamp is defined at module scope but was called as self._drop_timestamp(...) at four call sites (metrics calculation, TTFT training, TPOT training, and ensemble training) in training/training_server.py. This PR calls the module-level function directly at all four sites and removes the @staticmethod decorator on _drop_timestamp, since it has no effect on a module-level function.

Why is this change needed?

Each of the four call sites raised AttributeError: 'LatencyPredictor' object has no attribute '_drop_timestamp'. The error was caught by a broad except block, so the service kept running and reporting ready, but train() silently no-opped: the current model was never replaced and predictions stayed at the cold-start default of approximately 10.0 for both TTFT and TPOT.

How was this tested?

• Unit tests added/updated
• Integration/e2e tests added/updated
• Manual testing performed

Added tests/test_training_server.py, which calls _calculate_metrics_on_test with minimal synthetic data and asserts no error is logged. Verified locally that this test fails against the pre-fix code (it catches the AttributeError being logged) and passes with the fix. Also ran ruff check on both changed files with no findings.

Checklist

• Commits are signed off (git commit -s) per DCO
• Code follows project contributing guidelines
• Tests pass locally (make test)
• Linters pass (make lint)
• Documentation updated (if applicable)

Related Issues

Fixes #80