Use the XR_FB_HAND_TRACKING_MESH extension to render hand models when supported,
and fall back to the bundled GLB hand models otherwise. This fixes incorrect hand
model rendering on PFDM MR device, Have tested with PFDM MR V4.5.0 version.
The server only exports its D-Bus interface and sets JsonConfiguration
from inside on_name_acquired (server/main.cpp), after owning the
well-known name io.github.wivrn.Server. The dashboard's GetAll
triggered from on_server_dbus_registered can therefore race ahead of
that and come back with an error or an empty reply.
Previously the dashboard set ServerStatus to Started as soon as the
name was owned, so the rest of the UI assumed the server was usable
while JsonConfiguration was still missing. SettingsPage in particular
loaded its values once in Component.onCompleted and ended up showing
defaults (Encoder Auto, SteamVR checkbox unchecked, OpenVR / Autostart
blank), even though the configuration on disk was correct.
Move the Started transition into refresh_server_properties, only after
GetAll returns a reply that actually contains JsonConfiguration. If the
initial reply is errored or empty, retry every 200 ms until the server
has exported its interface. When the dashboard starts and finds the
service already registered, set Status::Starting in the constructor so
Main.qml does not see Stopped and spawn a duplicate process.
With this, ServerStatus::Started now consistently means 'server is
fully usable, properties are populated', and SettingsPage no longer
needs to handle the asynchronous arrival of JsonConfiguration itself.
Compositor release left layouts as eUndefined while encoder acquire
transitioned to eVideoEncodeSrcKHR (or eTransferSrcOptimal on the NVIDIA
tmp_image path); the spec forbids that layout mismatch for exclusive-mode
QFOT. NVIDIA honoured it by treating encode-queue contents as undefined
(solid green frame).
Encoders now publish target_layout. The compositor's release and the
encoder's acquire both transition eGeneral -> target_layout in a single
barrier; per VkImageMemoryBarrier2 the layout transition is performed
exactly once between the queues, so no second non-QFOT barrier is needed.
target_layout is eVideoEncodeSrcKHR on the direct vulkan encode path and
eGeneral elsewhere.
Also fix optimal_transfer(): the inner `auto to = ...` shadowed the
parameter, so the maintenance9 check was effectively always false.