Simplify Pixelbot API failure paths

[pierrebrunelle]

Pixelbot still carried duplicate exception wrappers, notification transports, and unused REST endpoints after the Pixeltable 0.7.7 upgrade. This follow-up removes those paths so the app has one sanitized failure boundary, one notification implementation, and a smaller supported API.

Changes:

• remove seven unused file/history/TTS/database endpoints and their orphaned models/types
• let unexpected route failures reach the application-level sanitized 500 handler while preserving explicit validation and not-found responses
• restrict transient retries to synchronous read-only operations so writes, provider calls, and notifications are not repeated after partial success
• share typed Slack, Discord, and webhook delivery across HTTP and Pixeltable tools; log only redacted destination origins
• reuse one isolated Pixeltable test catalog and stop its PostgreSQL process after the pytest session
• document why Torch and Transformers remain required by the core CLIP image/video indexes

Validation:

• uv run ruff format --check pixelbot tests
• uv run ruff check pixelbot tests
• uv run mypy pixelbot
• uv run pytest -q twice: 15 passed on both runs, with no test PostgreSQL process left running
• frontend lint, 5 Node tests, and production build; initial chunk 271.64 kB
• uv lock --check and uv build
• pxt schema check pixelbot/app.py
• schema diff against an applied isolated catalog: 19 unchanged, zero operations
• real pxt service update smoke test: health, memory, context info, and notification log returned 200; removed endpoints returned 404; service stopped cleanly

Provider calls were mocked; no paid provider or hosted deployment was run.