chore(deps): migrate pipecat integration to pipecat-ai 1.x (clears 4 high-severity alerts)

[dcbouius]

What

Migrates the pipecat integration from pipecat-ai>=0.0.100,<1.0 to >=1.4.0,<2.0, clearing four high-severity Dependabot alerts that have no fix available within the 0.0.x line:

Alert	Advisory
#1006, #1005	Telephony WebSocket /ws unauthenticated call-control abuse
#560, #559	Path traversal in Pipecat runner /files endpoint (arbitrary file read)

The fix versions (1.2.0 / 1.4.0) are only on the 1.x line, which the integration previously capped out (<1.0, "1.0+ restructured modules; needs migration"). This PR does that migration.

Why it's a small change

pipecat 1.x replaced the per-provider OpenAILLMContext with the universal LLMContext and removed the pipecat.processors.aggregators.openai_llm_context module. The processor already imported the modern LLMContextFrame, and LLMContext.messages is still a live list of OpenAI-format dicts — so the recall/retain/inject logic is unchanged. The migration is mostly swapping the removed legacy imports.

Changes

• memory.py — drop the now-impossible legacy OpenAILLMContextFrame import branch; match on LLMContextFrame directly. Runtime injection logic untouched.
• tests — build frames from LLMContextFrame; add TestRealLLMContext that exercises a real pipecat LLMContext + LLMContextFrame (not mocks) to pin the live-list-mutation contract the integration relies on — the migration's highest-risk assumption.
• examples — migrate basic_pipeline.py to LLMContext + LLMContextAggregatorPair + LLMRunFrame kickoff (create_context_aggregator / get_context_frame were removed in 1.x); update interactive_chat.py mock frame.
• pyproject — pipecat 1.x requires Python >=3.11, so bump requires-python and drop the 3.10 classifier (CI's .python-version is already 3.11).
• README / docstrings — OpenAILLMContextFrame → LLMContextFrame.

Testing

19 passed, 1 skipped (live)

uv run ruff check / format clean; uv lock --check consistent; wheel builds.