Measure routing impact, bundle hooks, preview team metrics

How we got here

The previous releases made routing advice and hook installation more visible, but they still left two operational questions open. First: does routing actually help inside a fixed 5h plan window, or does it only look cheaper in token equivalent dollars? Second: can a compiled ax binary install its hook SDK without depending on source files sitting beside it?

v0.35.0 answers both. The range adds a new `routing-impact` module and command family, then changes the build to generate and embed bundled hook assets. In parallel, Studio picked up a team metrics board and the installer became more graceful around IDE daemon state.

What changed

Routing gets an A/B receipt (#576): `ax routing impact` records paired work blocks with routing off and on, snapshots the Claude 5h quota window at each edge, and reports the result. The report keeps token-equivalent dollars as context, but the headline is plan-window utilization because that is the scarce resource on a fixed plan.

ax routing impact begin --arm=off --label=baseline
# work a comparable block
ax routing impact end

ax routing impact begin --arm=on --label=routed
# work the matched block
ax routing impact end

ax routing impact report

Compiled builds carry the SDK hooks (#574): the build now runs `scripts/gen-hooks-embed.ts` and includes a generated hook manifest, so `ax hooks` can install the bundled SDK hooks from the compiled binary. The hook guard names moved into a shared module, and the workspace tests cover both source and compiled layouts.

Studio gained a team metrics board (#567): the instrument shell now has a team metrics route and visualization layer, including the first `ax-for-teams` paywall shape. This is the early product surface for comparing team-level usage and outcomes rather than only local single-user runs.

Install and inspect paths were hardened: the IDE daemon model became doctor-aware and degrades more cleanly during install, while session inspect pagination stays anchored to the server window instead of drifting while a user pages through a large transcript.

Why it matters

Routing only earns a place in the workflow if it can prove it preserves scarce plan capacity without making the work worse. v0.35.0 gives that experiment a local receipt. At the same time, embedding hooks in the binary moves ax closer to a tool that can be installed and repaired from one artifact, which matters more as the hook layer becomes the front door for routing and guardrails.