feat(core): bump galileo-core to ^4.4.0, drop fallback ControlSpan tests (HYBIM-788)

[etserend]

Summary

Port of upstream commit fa93ab2 from rungalileo/galileo-python (#615) — third commit in the HYBIM-788 migration.

• Bumps galileo-core from ^4.3.0 → ^4.4.0 in both main deps and test group
• Runs poetry lock to update the lock file to galileo-core 4.4.0
• Replaces test_logged_trace_roundtrip_with_fallback_control_span and test_fallback_control_span_rejects_non_uuidish_id_fields with direct native equivalents — the builtins.__import__ + importlib.reload fallback machinery is dead code now that 4.4.0 ships native ControlSpan unconditionally
• Removes unused top-level imports (builtins, importlib, three splunk_ao.* module refs) from both test files

Namespace translation applied

No from galileo.* refs — this commit only touches pyproject.toml, poetry.lock, and two test files. galileo-core dep name is intentionally unchanged (external dep).

Test plan

• poetry run pytest tests/schemas/test_logged.py tests/test_agent_control_bridge.py -v — 49 passed
• poetry run ruff check src/splunk_ao/ --exclude src/splunk_ao/resources/ — clean
• grep -r "from galileo\." src/splunk_ao/ — no results

Part of

HYBIM-788 — commits applied in order: 3 of 3 (e42a569 → 6ccd06a → fa93ab2)

Stacked on: #74 (feat/HYBIM-788-port-6ccd06a — jsonl_flat)

🤖 Generated with Claude Code

[fercor-cisco]

Port fidelity review — nothing is being lost ✅

Compared upstream commit fa93ab2 against this PR line-by-line across all 4 files, synced the venv to galileo-core 4.4.0, and ran the affected suite (49 passed).

Per-file check

File	Upstream	Fork PR	Verdict
pyproject.toml	^4.3.0 → ^4.4.0 (main + test extras)	identical bump	✅
poetry.lock	galileo-core 4.3.0 → 4.4.0, wheel hash eb7995c…74e8	same version, byte-identical wheel hash	✅
tests/schemas/test_logged.py	2 fallback tests → native equivalents, drop builtins/importlib/3 module imports	same logic, correctly namespace-translated (galileo.* → splunk_ao.*)	✅
tests/test_agent_control_bridge.py	fallback test → native, drop same imports	same	✅

Specifically confirmed not lost

• Both renamed tests keep upstream's exact assertions (type(...) is LoggedControlSpan, ControlSpan.model_fields["name"].default, the 4-field parametrize).
• All removed imports (builtins, importlib, control_module/logger_module/logged_module) have zero lingering references — the remaining fake_agent_control_modules matches are an unrelated fixture.
• The PR correctly leaves src/splunk_ao/logger/control.py's fallback machinery in place — upstream also didn't touch source in this commit, only the tests. With 4.4.0, HAS_NATIVE_CONTROL_SPAN resolves to True (native ControlSpan lives in galileo_core.schemas.logging.span), so the tests exercise the native path exactly as intended.

Two deviations from upstream — both intentional and harmless

1. Import placement — upstream added a top-level from galileo.logger.control import ControlSpan; the fork uses function-local imports in both tests. Functionally equivalent. I verified the top-level style makes the fork's newer ruff (0.15.20) isort auto-reorder it, so the function-local choice cleanly sidesteps that.
2. Pre-existing fork baseline — the lock content-hash differs and pyproject shows ^3.11 (vs upstream ^3.10); unrelated prior fork state, not something this commit should carry over.

Note (not a defect)

The installed 4.4.0 wheel's in-package galileo_core.__version__ string still reads "3.96.0" (a stale constant baked into the wheel upstream). Dist metadata and the locked hash both correctly identify it as 4.4.0, and it's the identical artifact upstream pinned — an upstream cosmetic quirk, not a porting issue.

No changes needed — the port is complete and correct.

🤖 Generated with Claude Code

[fercor-cisco]

Approved but there is a merge conflict you will need to resolve.