feat(ipodtouch4): add persistent Pocket Runtime hot updates

[HalfSweet]

The iPod touch 4 host currently embeds each guest into its IPA. Add a separate PocketRuntime.app that accepts .pocket updates over the shared 3DS Runtime protocol, keeping native installation separate from guest iteration.

• Add bun ipodtouch4:runtime commands for native build/deployment, USB pairing, guest packaging, push, source watching, DevTools and status. USB uses the existing pinned SSH forwarder; LAN supports paired discovery and explicit addresses.
• Validate packages before replacing the guest, wait for its first GLES presentation, and commit active/last-good generations. Release the old QuickJS realm, GPU resources and touch identities between guests. Recover from guest failures and damaged stored packages, with embedded Clear as the final recovery guest.
• Keep the native listener available when a guest fails, bound QuickJS execution and socket work, and reconnect across foreground/background transitions. Share the wire codec and desktop client with 3DS while retaining compatibility entry points.
• Add real QuickJS/C ABI integration tests and run them in the Linux/macOS native harness workflow. Document the portrait viewport, reload semantics and deployment commands.

Validation:

• Built and signed the ARMv7 PocketRuntime.ipa against the pinned iOS 6 sysroot; built Clear's .pocket.
• Real QuickJS integration: uploads, pairing, fragmentation, interrupted transfers, malformed/foreign packages, eval/first-frame/later-frame failures, execution timeouts, restart recovery, compiled Clear rendering and DevTools queries.
• Legacy Apple/3DS regression tests, package/CLI tests, npm artifact checks, contract drift checks, workflow lint and git diff --check passed.
• Full tsc --noEmit reports two existing errors in unchanged tests/companion-session.test.ts (lines 19 and 26). Full-crate rustfmt reports existing formatting differences in the GLES files and a pre-existing test in lib.rs; the added Rust module passes formatting.

Device acceptance remains outstanding: this machine lacks libimobiledevice/iproxy and the iPod's pinned SSH credentials. Host tests do not exercise UIKit, GPU presentation or device installation/USB forwarding. This version uses a 320×480 density-2 foreground Runtime; screenshots continue through the existing USB capture command.

[doodlewind]

Before merging, I’d like to tighten a few architecture boundaries. The implementation direction is good, but this PR is large enough that I don’t want the iPod runtime to become a second runtime architecture that drifts from 3DS.

My merge gate would be:

1. Unify PKRT server semantics.
   We now have two independent PKRT server state machines: hosts/3ds/src/devserver.c and engine/runtime/dev_server.c. They already differ in protocol behavior (for example unknown frames / auth rejection behavior). Please move the common auth, framing, control, upload, timeout, and dispatch semantics into one shared implementation, leaving only platform/socket-specific integration in the hosts. Target name / discovery label should also be configuration, not hard-coded iPod-specific values in the generic runtime layer.

2. Move target/build-plan admission policy out of quickjs-c.
   pocket_runtime_validate_plan() currently hard-codes target policy such as presentation: "native" and the allowed feature set (input.touch, text.glyphs.baked). The QuickJS executor should not own device capability policy. Please keep the validate_plan host callback, but implement the actual validation in the package/target-contract layer or generated host-side contract code, so there is one source of truth for target capabilities.

3. Finish the generic runtime-client split.
   tools/pocket-runtime-client.ts should contain only generic PKRT client/session functionality. PICA RGB decoding, dual-screen composition, and PNG generation are 3DS-specific and should live in a 3DS-specific module. Also avoid production tooling importing tests/png.ts.

4. Do one real-device acceptance pass before merge.
   The host harness is strong, but it does not cover UIKit/EAGL lifecycle, actual presentation, installation, or USB forwarding. Please verify repeated guest replacement and recovery on a real iPod touch 4, including first-frame failure, later-frame failure, foreground/background transitions, reconnect, and several consecutive reloads. In particular, confirm the GL teardown/recreation path does not leak or retain stale resources.

I don’t think this needs a rewrite. The guest replacement/recovery manager and .pocket admission work are solid; the main issue is making sure the new persistent runtime becomes shared infrastructure rather than an iPod-specific fork of the existing 3DS runtime behavior.