host: fix source-mode worker startup churn (esbuild-bundle the worker entry once)

[brandonpayton]

What

In source checkouts (no host/dist/*.js), every guest process spawns a fresh tsx loader for its worker entry (~500ms bootstrap each). Homebrew forks/spawns hundreds of short-lived subprocesses per command (git, curl, ruby, shells), and under that concurrency the per-worker tsx path stalls — "source-mode worker startup churn." This was a real contributor to the brew config/doctor timeouts observed in the brew-in-Kandelo probes.

Fix

NodeWorkerAdapter gains a middle load path: when running from source, bundle the .ts worker entry once with esbuild into a single cached .mjs, and launch all workers from that prebuilt bundle. Load order becomes compiled-dist → esbuild-bundle-once → tsx-fallback. Only affects source-mode dev/probe runs; production/dist is unchanged. No guest-observable behavior change, no ABI change. Adds esbuild (devDependency) and a node-worker-adapter.test.ts.

Context

This is the one host fix from the brew-in-Kandelo frontier (originally on the kd-9jd worktree) not yet on main; every other scattered fix has already been independently merged. It's folded into the Track B2 brew-bootstrap re-verification branch. Opening standalone for review.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

[brandonpayton]

Verification: host/test/node-worker-adapter.test.ts passes locally under the dev-shell (Vitest 4, 1 test, 68ms). Full source-mode benefit is exercised by the Track B2 brew-bootstrap re-verification (many-worker Homebrew workload) on the same branch.