feat(session): render the EGFX graphics pipeline output

[Greg Lamberson (glamberson)]

Summary

• The ironrdp-egfx compositor exposes changed output regions via drain_output(), but nothing consumed them, so an EGFX session decoded frames and dropped them.
• Drains the compositor from ActiveStage::process: composites each completed-frame OutputUpdate into the DecodedImage and emits ActiveStageOutput::GraphicsUpdate, so every session consumer renders EGFX with no new code. No-op when the graphics DVC is not registered.
• Adds a by-type mutable DVC accessor DrdynvcClient::get_dvc_mut, mirrored on the session (the x224 processor and ActiveStage), matching the existing get_dvc.
• All regions drained in one pass are composited first by composite_graphics_updates, then surfaced as a single GraphicsUpdate covering their union. A consumer may redraw whatever region an update names, and ironrdp-client rebuilds the whole framebuffer for each one, so emitting per region would copy the desktop once per rectangle. A single RDPGFX_SOLIDFILL_PDU or RDPGFX_CACHE_TO_SURFACE_PDU can name up to u16::MAX of them.

Validation

• cargo xtask check fmt/lints/tests/typos/locks all pass, including the dependency guard.
• Four tests cover the coalescing: two disjoint deltas collapsing to the rectangle spanning both, 64 deltas yielding exactly one region, a single delta passing through unwidened, and an empty drain surfacing nothing. Checked against a reverted coalescing, where the first two fail.
• The loop lives in composite_graphics_updates rather than inline in ActiveStage::process so it can be tested without standing up an x224 processor. It takes (ExclusiveRectangle, Vec<u8>) rather than OutputUpdate, since that type is #[non_exhaustive] and cannot be constructed from ironrdp-session; this keeps the change inside the crate instead of widening an already-merged API for testability.

Notes

• ironrdp-session already depends on ironrdp-displaycontrol and ironrdp-graphics; ironrdp-egfx is consumed the same way. The guard keeping ironrdp-connector and sspi out of ironrdp-session still passes.
• Reuses apply_rgba32 (previously gated behind the qoi feature); converts the compositor's exclusive-rectangle regions to the session's inclusive-rectangle convention.
• fix(dvc)!: replace DVC wrappers with typed accessors #1377 and feat(egfx): composite client surface commands into pixel buffers #1460 have both merged, so this now sits directly on master and the diff is its own change alone: 6 files, +160/-6.
• feat(client): register the graphics pipeline dynamic virtual channel #1462 is stacked on this one, so the merge order is this then feat(client): register the graphics pipeline dynamic virtual channel #1462.
• Part of [tracking] Client-side Graphics Pipeline (EGFX): wire the existing decode core into the connector and client stack #1464. Motivated by Client can't connect to GNOME Remote Desktop: no Graphics Pipeline (EGFX) advertised — the blocker after #1371 #1446.

[Benoît Cortier (CBenoit)]

This high-risk, protocol-relevant EGFX/DVC integration is additive and feature-gated, with no valid findings anchorable to changed files; the protocol review identified valuable follow-up tests but no reportable defect, so it is ready for human review.