fix(pc-list): surface custom HTTP/HTTPS ports in UI - #125

…unnel (#132)

* chore(deps): bump moonlight-common-c to 31a2a45

Pulls in remote text context updates (#28), opt-in dynamic HDR wire
constants (#26) and received-video-bytes stats (#27). The new
RemoteTextContextStream.c joins the CMake source list (missing it
breaks the final link with an undefined decodeRemoteTextContextPacket).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore(release): changelog and version catch-up for 1.0.0.807-811

Backfills release notes for #121/#123/#125-128 that shipped without
CHANGELOG entries in this checkout.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* build: drop pinned @ohos/hvigor dependencies

Pinned dependencies make hvigor install a second local instance next
to the DevEco Studio wrapper, splitting plugin resolution into two
hvigor instances (00302013 "root node is not yet available"). The
dependencies key stays as an empty object: it is required by the
hvigor-config schema.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(input): OTG gamepad passthrough to Sunshine via USB/IP reverse tunnel

Forward an OTG-attached gamepad to the Sunshine host as a native USB
device over the same reverse tunnel contract as moonlight-qt and
moonlight-android (TLS with paired client cert, pinned server cert,
one-shot token from /api/v1/usb-forwarding, then opaque USB/IP bytes).

Native:
- usbip_server: in-app USB/IP 1.1.1 server on loopback over the USB
  DDK. Devices are registered from usbManager as (busNum, devAddress)
  using the official deviceId encoding (busNum << 32 | devAddress),
  since OH_Usb_GetDevices() is empty for normal apps. Full
  usbip_usb_device in DEVLIST/IMPORT replies, endpoint-scoped interface
  handles, and IN URBs driven in 200ms DDK timeout slices so a pending
  read never blocks unlink or queued control/OUT traffic. The listener
  accepts only the loopback source port the tunnel pre-bound, so no
  other local process can drive the exported device.
- usbip_tunnel: reverse tunnel client ported from moonlight-qt
  (VerifyNone + manual DER pin compare, TCP_NODELAY, bounded startup
  handshake, reason surfaced from refused forwards).
- usbip_napi: UsbIp NAPI object wiring server + tunnel singletons and
  the port authorization handoff.

ArkTS:
- UsbForwardingService: stream-scoped orchestration - exclude the
  device from the local USB driver, request rights, register it, fetch
  the capability, read pairing certs, start the tunnel; on failure
  release back to the local driver. v1 forwards one device.
- Stream menu entry with live status (disabled/no-device/starting/
  ready/error) and start/stop/retry actions; settings toggle
  "USB 直通主机（实验）".

v1 limits: one device per session, no isochronous endpoints (DDK does
not expose them). Requires Sunshine with USB forwarding enabled and
usbip-win2 installed on the host.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(usbip): address review findings on PR #132

- usbip_server: RET_SUBMIT/RET_UNLINK headers were appended past the
  pre-sized 48-byte buffer instead of written at protocol offsets, so
  every URB reply was malformed on the wire. Add storeU32/storeI32 and
  write fields at their fixed offsets.
- usbip_server: publish the accepted client fd and shutdown() it in
  Stop(), so a handler parked on a half-PDU read from a stalled peer
  cannot wedge the accept-thread join.
- usbip_napi: guard tunnel state/message with a dedicated mutex (the
  tunnel thread writes them while JS threads query; g_mutex may be
  held across the joining Stop); free the tsfn event when a
  nonblocking enqueue fails, as ownership stays with the caller.
- usbip_tunnel: resolve the Sunshine endpoint with getaddrinfo so
  IPv6 literals and hostnames connect instead of silently targeting
  255.255.255.255; publish fds as -1 before close() at every cleanup
  site so a concurrent WakeSockets() cannot shutdown a recycled fd.
- UsbForwardingService: reclaim leftover resources at begin() entry -
  a superseded flow skips teardown on its generation guard.
- CHANGELOG: qualify the passthrough feature wording (no isochronous
  endpoints yet, hardware validation in progress).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(usbip): define PDU header offsets used by RET_SUBMIT/RET_UNLINK

The review fix referenced kStartFrameOffset/kNumPacketsOrErrorOffset/
kErrorCountOffset, which were dropped from the constants block during
the server rewrite - CI (and any real compile) failed with undeclared
identifiers. Restore them (offset 36 is interval in CMD_SUBMIT and
error_count in RET_SUBMIT) and drop the now-unused appendI32.

The earlier local verification missed this: the build ran with the
nativelib dependency resolving through a stale oh_modules junction into
the main checkout (a different branch's tree) and empty submodules, so
the worktree native build never compiled this file. This build compiles
usbip_server.cpp for both ABIs from the worktree's own sources.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(usbip): address round-2 review findings on PR #132

- runHeadInUrb: a MSG_PEEK returning fewer than 48 bytes is a partial
  PDU from TCP segmentation, not a dead connection. Only EOF and hard
  socket errors close the drive; partial headers hand off to the main
  loop's blocking readAll, and EAGAIN continues the device wait.
- Stop/AcceptLoop: publish the accepted client under a lifecycle mutex
  paired with Stop()'s teardown, so a connection accepted just before
  teardown can never be published after Stop's shutdown sweep (which
  would leave join() waiting for the session to end on its own).
- LoadDdk: require the control-transfer pair and DestroyDeviceMemMap,
  and reject a torn config-descriptor pair (exactly one present);
  previously SendControlRead/WriteRequest were dereferenced unchecked
  and FreeConfigDescriptor was called unguarded.
- sendRetSubmit: allocate payload bytes only when data is provided;
  the header still reports actual_length (OUT replies carry no
  payload).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(usbip): RET_UNLINK reports -ECONNRESET for canceled pending URBs

Both CMD_UNLINK paths (the timeout-slice peek in runHeadInUrb and
processPdu) replied status 0 when the target IN URB was still queued,
which the host reads as a successfully completed URB. A URB canceled
before RET_SUBMIT carries -ECONNRESET, matching what a real USB stack
reports for unlinked URBs; URBs no longer pending keep the ENOENT
reply.