fix: wire ProvenaCrewListener to current CrewAI event bus API

[Sharvari713]

What does this PR do?

ProvenaCrewListener never received real events — three separate problems
compounded into a silent no-op: it imported from a module path CrewAI no
longer has, it never implemented setup_listeners() (the method CrewAI's
BaseEventListener requires every subclass to define, so the bare
on_tool_usage_finished/on_agent_execution_completed methods were never
invoked by anything), and it read a nonexistent event.agent_name. A user
following the README's exact usage got zero trail records from a real
crew.kickoff() run, with no error anywhere.

The fix:

• Switched imports from crewai.utilities.events.* to crewai.events
• Implemented setup_listeners(), registering ToolUsageFinishedEvent and
  AgentExecutionCompletedEvent handlers via @crewai_event_bus.on(...)
• Read agent identity via event.agent.role instead of the nonexistent
  event.agent_name

Verified end to end against a real installed crewai (1.15.22): constructed
real event objects, fired them through the real crewai_event_bus, confirmed
trail records appear with the correct source_name.

Also bumped the crewai extra's minimum from >=0.80 to >=1.0.0.
Confirmed by direct install/inspection in a throwaway venv, not a guess:
crewai 0.80.0 through 0.98.0 have neither the old nor the new event API
at all — they use a third, unrelated Emitter/emit/on decorator system.
The class-based event API this adapter depends on first appears at 1.0.0,
with the same fields (agent, output, tool_name) present through
1.15.22. crewai's own metadata pins <3.14,>=3.10 at both versions,
consistent with provena's own >=3.10 floor — no change needed there.

Rewrote tests/test_adapters.py's CrewAI fixture to fake the real API shape
(crewai.events, setup_listeners()/bus .on() registration) instead of
the old wrong module path, and rewrote all tests to fire through that
registration path instead of calling the old unbound methods directly.
Verified: all tests pass with crewai genuinely uninstalled, and all fail
against the reverted buggy adapter code (it can't even import successfully
under a correctly faked module layout — falls straight to the ImportError
placeholder).

Checklist

• Tests added/updated for the change
• ruff check src/ tests/ passes
• ruff format --check src/ tests/ passes
• mypy src/provena/ passes (1 pre-existing unrelated error on main, untouched by this change)
• pytest passes with no failures (575 passed, 35 skipped — unrelated optional integrations)
• CHANGELOG.md updated

Related Issues

Fixes #193