DisplayDriver seam: fold the bulge-exclude span into present() (#345)

[timohueser]

Closes #345 (epic #353, phase 2).

What

1. Trait: DisplayDriver::present(&mut self, exclude: Option<(u16, u16)>) -> bool — excluded rows are diffed (the store tracks the clean fb) but not pushed; the overlay plane owns repainting them. Contract documented on the trait.
2. Shared skeleton hoisted: the diff → clip-around-exclude → whole-frame-minus-exclude-on-span-overflow logic moves from Ls021Flpr::present_within to obc_platform::RowDiff::diff_clipped(fb, stride, exclude, &mut spans) -> &[(u16,u16)] (caller-provided scratch, returns the filled prefix — no_std, no alloc). Four new host tests: clipping, store-updated-for-excluded-rows (a later un-excluded present doesn't re-push), priming-with-exclude, and the overflow fallback respecting the exclude.
3. FLPR: present_within shrinks to diff_clipped + push_spans and goes pub(crate) (only the display::ls021_flpr adapter calls it); the map plane's match overlay_span + concrete present_within call is deleted — it calls the seam's present(overlay_span) unconditionally.
4. ST7789: present(exclude) bands only the clipped spans — the documented "dev-only-best-effort" mid-hold flash is gone. The tft MapDisplay now holds the InputPlane handle: poll_overlay returns the live bulge span (read-only overlay_rows — the one-shot take_overlay_dirty edge stays with input_overlay_task, which still owns the bulge push), and hold_progress() returns the real value, so the in-screen Reset-bar fill finally moves on tft.
5. Convergence: poll_overlay/hold_progress/render_present bodies are now near-identical across the two MapDisplays; the structs stay separate (bus-mutex vs owned-panel is the genuine asymmetry). Not folded further per the issue ("full merge optional stretch").

Note: render_present's FLPR-only reset_diff() fault-recovery call remains a concrete-type touch — that recovery shape is #349's remit.

Verification

• 4 new diff_clipped host tests + all existing rowdiff/clip_span tests pass (workspace cargo test green).
• grep present_within hits only ls021_flpr.rs + its seam adapter; the map plane has no concrete present call left.
• All six board feature shapes build; clippy -D warnings-clean on default/tft/ble.
• On-glass checks pending (no DK on my end): FLPR bulge-during-redraw parity, and tft mid-hold no-blank + Reset-bar fill.

Docs

docs/content/software/rendering.md credited only the FLPR with presenting around the bulge — updated in a separate docs: commit (link check passes). firmware/docs/ls021-flpr.md doesn't describe the present seam; the public display-protocol page is wire-level only — both unaffected.

🤖 Generated with Claude Code