fix(cli): host fingerprint v2 — survive Mac restarts (1.34.17)

v1's sha256(host_id || mac) used the lex-first non-virtual interface's
MAC — usually en0 on Wi-Fi Macs, whose MAC Apple's privacy feature
re-randomizes across reboots. After a restart the recomputed hash no
longer matched the stored one and the daemon entered a launchd
respawn loop until manual `claudemesh daemon accept-host`.

v2 reads IOPlatformUUID via ioreg on macOS (burned into EFI, stable),
rejects locally-administered MACs in the picker, extends the ignored-
interface list with anpi/bridge/ap[N], and prepends "v2\0" to the
hash so v1 and v2 hashes can never collide on the same inputs.

Migration is silent: a stored v1 fingerprint that still matches under
the v1 algorithm is transparently rewritten as v2 with no error; v1
stores that fail v1 are reported as genuine mismatches as before;
unknown future schema_versions return `unavailable` without
overwriting.

Drive-by fixes for two pre-existing test-infra papercuts found while
validating: turbo's `test` task now depends on `build`, and a new
vitest globalSetup rebuilds the CLI on demand with ~/.bun/bin and
Homebrew layered into PATH — golden tests (whoami, --version) no
longer fail opaquely after a clean checkout.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudemesh-cli",
-  "version": "1.34.16",
+  "version": "1.34.17",
   "description": "Peer mesh for Claude Code sessions — CLI + MCP server.",
   "keywords": [
     "claude-code",