Removed OSMesa, replaced with BUILD_BACKEND_HEADLESS_SOFTWARE

[kerberjg]

OSMesa is removed from Mesa 25.1+ and no longer a viable headless rendering path.

This PR replaces the headless EGL backend (which required libosmesa6, libgles2, and libegl1) with a zero-dependency headless mode built on the existing SoftwareBackend (Flutter kSoftware renderer, NoneSink by default).

New CMake flag: BUILD_BACKEND_HEADLESS_SOFTWARE

• Mutually exclusive with all other backends (same as BUILD_BACKEND_SOFTWARE)
• Forces BUILD_SOFTWARE_SINK_DRM, BUILD_SOFTWARE_SINK_FBDEV, and BUILD_SOFTWARE_INPUT_LIBINPUT OFF; no system library deps required
• No OSMesa/EGL/Wayland linkage: IVI_HAVE_EGL and waypp include gates no longer include the headless case; waypp subproject is not built

SoftwareBackend gains a const bool is_headless constructor parameter:

• Ctor: rejects hardware sinks (NativeSize() != nullopt) in headless mode, swapping them for NoneSink with a warning
• GetVsyncCallback(): returns nullptr in headless mode (no vblank source)
• SetCursor(): no-op in headless mode (no display to composite onto)
• CreateSurface(): trace-logs the no-op in headless mode
• New GetSink() accessor exposes the active sink for in-process pixel capture (used by the reworked getViewRenderBuf API)

flutter_view.cc constructs SoftwareBackend with is_headless=true under BUILD_BACKEND_HEADLESS_SOFTWARE, using MakeSinkFromEnv() so test fixtures can set IVI_SW_SINK=memory to enable pixel capture without recompiling.

App::getViewRenderBuf() is reimplemented via MemorySink::SnapshotLatest(), returning std::vector<uint8_t> with row_bytes/height out-params instead of a raw GLubyte* into an OSMesa framebuffer.

Also - all references to OSMesa are removed.

[jwinarske]

@kerberjg There seems to be some confusion.

e.g.

// In headless mode hardware sinks (DRM dumb / fbdev) are meaningless —
// there is no display to scan out to. If the caller passed one (e.g. via
// a stale IVI_SW_SINK env var), swap it for a NoneSink so the process
// never tries to open /dev/dri/* or /dev/fb*.

These are the software sinks as documented in shell/backend/software/README.md

Supported Software Sinks

Sink	Spec syntax	What it does	Vsync
NoneSink	none (default)	Discards every frame. CI engine-only smoke.	—
MemorySink	memory	Mutex-guarded std::vector<uint8_t> snapshot of the most recent frame. SnapshotLatest(&row_bytes, &height) exposes it to in-process test fixtures.	—
FileSink	file:<path-pattern>	Writes each frame as a NetPBM PAM (P7) file. Pattern with %d / %05d interpolates the frame index. Pattern without % writes the first frame only. Parent directories auto-created.	—
FbDevSink	fbdev[:<device>]	Opens /dev/fb0 (or operator path), mmaps, packs each Present. Auto-detects 32-bpp BGRA/BGRX or 16-bpp RGB565 from the driver's FBIOGET_VSCREENINFO offsets; refuses palettized / nonstd != 0 / other layouts with a clear error.	—
DrmDumbSink	drm-dumb[:<device>]	Opens /dev/dri/card0, picks first connected connector + its CRTC + preferred mode, allocates 2 dumb buffers (XRGB8888 by default, RGB565 via IVI_SW_DRM_FORMAT=rgb565 when the plane advertises it), modesets onto buffer 0. Per-frame: pack into the back buffer, drmModePageFlip. PAGE_FLIP_EVENT drives Flutter's vsync_callback with the kernel-provided scanout timestamp.	✓

Hold off on this PR until further notice. Major changes are incoming.

[kerberjg]

@jwinarske Yeah, that looks like something my Claude spat out. Thanks for the correction!

Please note this one is blocking for #239

[jwinarske]

closing in favor of #254