Stabilize missing-method instrumentation test across runtimes

[Copilot]

The Windows Python 3.9 failure came from a brittle test assertion in tests/instrumentation/base_tests.py: it expected runtime-generated AttributeError text that varies across Python/wrapt/platform combinations. This change updates the test to assert the stable agent-generated log message for skipped instrumentation instead.

• Problem

  • test_instrument_nonexisting_method asserted on "has no attribute", which is not a stable contract of the agent and differs on Windows/Python 3.9.

• Change

  • Update the assertion to match the stable log prefix emitted by AbstractInstrumentedModule.instrument() when a target method cannot be resolved.
  • Preserve the logger check against elasticapm.instrument.

• Effect

  • The test now verifies the intended behavior — skipped instrumentation of logging.Logger.non_existing_method — without depending on platform-specific exception wording.
  • No production error handling or workflow behavior is changed.

assert_any_record_contains(
    caplog.records,
    "Skipping instrumentation of logging.Logger.non_existing_method:",
    "elasticapm.instrument",
)

Fixes #2733