Fix local stability issues in startup, reports, chat, and prompts

[veil-chow-fyaic]

Summary

• reuse an already healthy backend instead of treating port 1733 as a fatal startup error, and avoid stopping a backend this Electron process did not spawn
• make daily reports deterministic by using the report start date, normalizing fenced Markdown, and updating an existing daily report for that date
• isolate chat agents per request to avoid shared event queue/session state across conversations
• harden prompt persistence and prompt lookup by saving user prompts under the writable user settings directory and falling back from screenshot_analyze to the legacy screenshot_contextual_batch prompt
• coerce LLM-returned extracted fields into stable shapes so list/dict summaries do not fail ExtractedData validation
• reduce noisy development update checks and improve notification content for generated tips

Related issues

Refs #365, #350, #342, #338, #318, #313, #253, #246, #231, #221, #174, #166.

Tests

• /Users/fuyo-aic/Projects/MineContext-fork-fixes/.venv/bin/python -m unittest discover -s tests -v
• python3 -m py_compile opencontext/context_consumption/generation/generation_report.py
• pnpm exec tsc --noEmit -p tsconfig.node.json --composite false
• pnpm exec tsc --noEmit -p tsconfig.web.json --composite false

Notes

This PR keeps the fixes focused on stability and backward-compatible fallbacks. It does not attempt to solve platform-specific capture limitations such as minimized-window capture on Windows.

[codeCraft-Ritik]

Solid maintenance-focused PR. There are several independent fixes here, but they all seem aligned around improving reliability and reducing unexpected behavior. The daily report determinism changes are particularly nice for consistency and debugging.