fix: address round-4 review feedback on semantic memory

[Dhirenderchoudhary]

Description

Addresses comments of the semantic memory feature (#619): #619 (comment)

Revised merge bar, all four items:

• /memory accept <n> re-indexing bug fixed — accepted indices are now tracked in a Set instead of splicing the proposal array, so accepting one proposal no longer shifts the numbers under proposals still on screen.
• Docs settings-path contradiction fixed — both docs now correctly say /settings → Advanced → Semantic Memory.
• Token budget wired to preferences — resolveProjectContextPreferences() is the single source of truth, wired into all three call sites (TUI, ACP agent, plain shell), plus two new cyclable rows in the Advanced settings panel.
• /memory list short IDs — memory.ts rewritten as a themed memory.tsx Ink component (matching tasks.tsx/mcp.tsx/agents.tsx), with 8-char short IDs and unambiguous-prefix resolution replacing raw 121-char UUID lines.

Also hardened the assistant-reversal detector per the reviewer's own follow-up: added a real contradiction check between assistant turns (isContradiction), widened the technical-evidence guard so a bare / no longer disables detection, and made the walk-back skip tool-call-only turns instead of bailing on any intervening assistant message. Plus: fenced project-context content can no longer be escaped by backticks in memory content, /memory propose output is now bounded (last 40 messages scanned, 20 proposals max), and the semanticMemoryEnabled default no longer lives in three separate places.

Closes #619

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update

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

Ran tsc --noEmit (clean), biome check on every touched file (clean), and targeted ava runs covering every touched spec file — 142+ tests passing, 0 failing.

Manual Testing

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

Checklist

• If this was for an open issue, I was assigned to it
• 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) — not applicable, no new logging paths in this change