fix: address review feedback on semantic memory

[Dhirenderchoudhary]

Description

• Retrieval: added a stopword list and a minimum relevance-ratio threshold to findRelevantMemories so incidental word overlap (e.g. "the", "in") no longer scores as relevant
• /memory propose: capped candidates per message and truncated evidence to snippets instead of full messages; explicit user statements can now be proposed even without matching one of the five keyword categories, while uncategorized assistant chatter is still filtered out
• Added /memory accept <n> to save a proposal directly instead of retyping it into /remember
• Wired project-memory recall into --plain and --acp, matching what the TUI already did (previously memories saved on those surfaces were never recalled)
• The semanticMemoryEnabled setting now gates writes (/remember, /memory accept), not just recall
• Added docs/features/semantic-memory.md and /remember//memory entries in commands.md
• Non-blocking cleanups: fixed a read-modify-write race in SemanticMemoryManager, fixed the token-budget loop silently reordering memories by size, fenced the injected memory content in the system prompt, removed a dead MemorySourceType value, fixed an order-dependent warning on accepted proposals

Closes #619

Type of Change

• Bug fix

Changeset

• Added a changeset (pnpm changeset) describing this change for the changelog

Testing

Automated Tests

• New features include passing tests in .spec.ts/tsx files
• All existing tests pass (pnpm test:all completes successfully)
• Tests cover both success and error scenarios

Manual Testing

• Tested with Ollama
• Tested with OpenRouter
• Tested with OpenAI-compatible API
• Tested MCP integration (if applicable)

Checklist

• Code follows project style guidelines
• Self-review completed
• Documentation updated (if needed)
• No breaking changes (or clearly documented)
• Appropriate logging added using structured logging (see CONTRIBUTING.md)