The PasteInterceptor processes the entire paste (including end marker)
before readline fires any line events. This caused multi-line pastes
to be submitted line-by-line because by the time the first line event
fired, the paste was already 'ended'.

Fix: Track expected line count when paste ends, and only flush after
all expected lines have been received.

Completely rewrites paste handling to work like Claude Code:
- Capture entire paste content in PasteInterceptor
- Display '[pasted N lines, M chars]' summary
- Store paste for consumption by line handler
- Emit newline to trigger readline submission

This fixes multi-line paste being submitted line-by-line.

Added /log and /history as standalone aliases for /git log,
matching the pattern of other git command aliases like /commit,
/branch, and /pr.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Added TypeDoc API documentation generation (pnpm docs)
- Created CHANGELOG.md from GitHub release notes
- Fixed docs/tools.md - removed outdated paste code examples
- Added ollama-cloud mentions alongside ollama in:
  - marketing/social-posts.md
  - docs/index.html
  - .github/ISSUE_TEMPLATE/bug_report.md
- Added docs/api/ to .gitignore (generated)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Clarified that AI agents must review PRs before merging:
- Added explicit warning not to merge without reviewing
- Added specific instructions for AI agents
- Added step to document review with comment
- Emphasized "merge only after review is complete"

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Changed default baseUrl from localhost to https://ollama.com
- Added OLLAMA_API_KEY support for authentication
- Added Authorization header to all API calls when key is set
- Updated docs to show ollama signin and API key options
- Removed incorrect OLLAMA_BASE_URL instructions

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Changed auto-detection to prefer Ollama (free/local) over OpenAI.
Users can still explicitly set embeddingProvider: 'openai' in config
if they prefer OpenAI embeddings.

Note: Anthropic does not offer embedding models - they recommend
Voyage AI as their preferred embedding partner.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>