Wizard onboarding: what's new, and how to flag issues

[VGIL77]

The onboarding wizard has picked up several enhancements over the last ~10 days. Posting so anyone who ran it before this week knows what changed, and anyone who's hit a snag has a clean way to report it.

What's new

1. Dedicated P2P onboarding step (src/wizard/onboarding.p2p.ts). Probes for the orchestrator binary across the 4 resolution paths, resolves _dnsaddr.p2p.bitterbot.ai with live peer count, TCP-probes the fallback bootstrap peer, auto-writes desktop/.env with the gateway URL and token so the Control UI connects on first pnpm dev.
2. Memory embeddings provider step. Configure OpenAI (or swap to a local alternative) during first-run instead of hand-editing config later. (0d0a507)
3. Web search provider step. Same pattern as above. (f0fb4dd)
4. tsdown clean race fix on pnpm dev:all. The build's clean phase was deleting files while the watcher was still spawning. Fixed. (c8abd22)
5. Expanded doctor subcommands. doctor-runtime, doctor-identity, doctor-skills, doctor-channels, doctor-web-search cover much more ground. Run bitterbot doctor any time something feels off.

Known rough edges

Tracking in issue #9. Summary of what we're still smoothing:

• Race between the tsdown clean phase and pnpm dev:all on slower disks (mostly fixed in c8abd22, watching for edge cases)
• Repetitive provider-key prompts on re-runs (planning a "skip if already configured" pass)
• P2P step on locked-down networks doesn't surface a clear next step when DNS bootstrap fails
• Orchestrator binary probe running before cargo build completes on very fresh clones

Hit a snag? Drop a comment using this template

The more specific, the faster we can cut a fix.

**OS + version:**            (e.g. macOS 14.4, Ubuntu 22.04, Windows 11 23H2)
**Node version:**            (output of `node --version`)
**Fresh clone or re-run?:**  (fresh / re-run of existing setup)
**Step you were on:**        (gateway, memory embeddings, web search, p2p, finalize, other)
**What happened:**
  <short description>

**Full terminal output from the point it broke:**
```
<paste here>
```

**`bitterbot doctor` output (relevant section):**
```
<paste here, if it flags anything red>
```

**Anything else odd in the environment:**
  (corporate network, VPN, WSL, custom Node path, etc.)

We'll fold fixes into issue #9 and close it once the rough edges are gone. Thanks for helping us shake this out.

[A561988]

My friend hit an issue on their Mac last week where the DNS bootstrap was silent on a restricted network, so having bitterbot doctor now to actually flag the specific resolution failure is great - thanks.

Does the wizard now support a local Ollama or LM Studio endpoint out of the box, or are we still defaulting to an OpenAI key first? For the privacy-conscious crowd (and to really show off the local-first mesh), making that local-swap as seamless as possible during the first-run would be a massive win for node velocity.

Great work on the tsdown race fix—that was definitely a 'rough edge' on slower SSDs.

[OhhJJ]

Let me know if you run into any problems, but your manuals are very helpful.