docs: vault search redesign design doc

[logancyang]

Summary

Replace the monolithic localSearch tool with three focused tools that mirror Unix find / grep / cat:

Tool	Analogy	Purpose
vaultFind	find	List files by metadata (time, tags, folder, title). Returns metadata + headings, no content. Always local.
vaultSearch	grep	Search file contents for a topic. Returns matching passages. Delegates to Miyo/semantic/lexical.
readNote	cat	Read full note content. Already exists, unchanged.

Motivation

The current localSearch handles time filtering, tag matching, title matching, AND content search in one tool. This caused #2267 where timeRange was silently dropped in the Miyo code path. More fundamentally, the monolithic design prevents the agent from composing multi-step queries intelligently.

Key Design Decisions

• vaultFind is always local — uses Obsidian's vault API and metadataCache. No Miyo, no embeddings, no Node.js dependency (mobile-safe).
• vaultSearch has NO metadata filtering — pure content search. Time/tag/folder filtering is vaultFind's job.
• Composition via paths parameter — vaultSearch accepts optional paths from a vaultFind result to scope content search to specific files.
• Headings from metadataCache — vaultFind returns headings (free from cache), giving the agent enough context to decide what to read without blowing up context.
• Architectural invariant: Time/tag/folder filtering is ALWAYS local (vaultFind). Content search ALWAYS delegates to best backend (vaultSearch). These never cross.

Query Examples

"What did I do last week?"          → vaultFind(timeRange=...)
"Notes about machine learning"      → vaultSearch(query="machine learning")
"Notes about X from last week"      → vaultFind(timeRange=...) → vaultSearch(query="X", paths=[...])
"Summarize Q3 meeting notes"        → vaultFind(folder="Q3Meetings") → readNote each
"Meeting notes mentioning John"     → vaultFind(tags=["#meeting"]) → vaultSearch(query="John", paths=[...])

Design Doc

Full design doc with schemas, 10 query examples, agent instructions, migration plan, and implementation notes:
designdocs/VAULT_SEARCH_REDESIGN.md