perf: narrow first startup pass to startup-critical data only

[justrach]

Goal

Keep the first startup pass as narrow as possible: collect only what startup immediately needs, and defer richer structures until later.

Why

We already know the gap with fff is mostly time-to-search-ready vs full code-intelligence startup. The first pass should optimize for readiness, not completeness.

Scope

• Define exactly which structures are required for:
  • tree
  • outline
  • find
  • hot
  • first search-ready path
• Defer non-critical structures until after startup readiness.
• Make the startup contract explicit in status/reporting.
• Preserve the fast warm mmap-backed search path.

Acceptance Criteria

• Startup-critical commands become available before the full detailed index finishes.
• Deferred work does not block the first user-visible tool call.
• Warm search path remains on par with current main.
• Metadata-only commands do not eagerly materialize unnecessary content/search state.

Regression Checks

Add explicit tests and benchmarks:

• Status test proving startup-ready and detailed-ready are distinct states.
• First-query MCP search benchmark immediately after startup on a large repo.
• Warm repeated MCP search benchmark after readiness.
• Cold tree benchmark with fresh HOME.

Merge Gate

Do not merge if any of the following regress beyond threshold versus main:

• Warm first MCP search: > 5%
• Warm repeated MCP search: > 5%
• Cold tree wall time: > 5%
• Result correctness / parity: any regression is blocking

Notes

This issue should not reintroduce the earlier dead-end path where deferred detailed work starved or blocked the first search.