video: show early hi-res DDF content inside an early display window

[LinuxJedi]

Fixes #186 (ECS Denise PAL/NTSC horizontal offset: bitplanes shifted left relative to sprites on overscan screens).

Root cause

The early hi-res DDFSTRT pre-fetch snap (af42e2d) unconditionally skipped the whole early-fetch width at the picture origin, assuming the display window always opens at the standard $81 edge. KS 3.2 programs ECS extreme-overscan screens with both an early fetch and an early window (Overscan editor edit display: DDFSTRT=$28, DIWSTRT h=$5D, DIWSTOP h=$1C7 via DIWHIGH, BPLCON1=$44). On real hardware those early-fetched words are inside the window; the snap discarded 80 native px instead of the physical 18, shifting the whole bitmap left with a blank colour-0 band at the right window edge while sprites stayed beam-anchored - exactly the reported symptom, on PAL and NTSC alike.

Fix

Remove the snap. The physical origin term (display shift minus DDF shift plus the framebuffer-origin clamp) already clips exactly the part of the fetch that lands left of the window: XSysInfo's DDFSTRT=$38 panel still skips its one pre-fetch word. The snap only ever differed from that term for windows left of the fetch reference - it was compensating for the old $83 hi-res fetch reference and became a no-op for standard windows when the reference was recalibrated to $81 (PR #106/#107 era).

Verification

• Full unit suite green (1434 tests); early_ddf_hires_origin_clips_prefetch_against_the_window_edge keeps the original XSysInfo/standard/KS-boot assertions unchanged (16/0/24) and gains an ECS 8375 overscan case asserting origin 18.
• All 20 golden render probes pass unchanged (no drift in the calibrated DDF placement web); clippy and fmt clean.
• End-to-end on the KS 3.2 Overscan editor's max-overscan edit screen (A3000 profile, ECS): picture now fills the window symmetrically on PAL and NTSC, matching the A4000/AGA rendering of the same screen; clicking Cancel at its drawn position closes the screen, so the bitmap and Intuition's hit-testing coincide again.
• ECS standard (non-overscan) Workbench screen and the AGA edit screen are byte-identical before/after the change; only the broken case moves.

docs/internals/video.md records the rule: hi-res early DDF stays beam-anchored and is hidden by the window comparator alone.

[Copilot]

Pull request overview

This PR fixes ECS Denise hi-res early-DDF rendering on extreme-overscan screens where the display window opens early, preventing an incorrect left-shift of bitplane content relative to sprites by removing an unconditional “snap” that discarded early-fetched words even when they are inside the window.

Changes:

• Removes the hi-res/SHRES FMODE=0 early-DDF “snap to fetch-word boundary” from ControlState::fetch_origin_native_shift, relying on the existing origin/window math to clip only what’s truly off-screen.
• Updates/extends the bitplane origin unit test to cover the KS 3.2 ECS extreme-overscan case from issue #186 while preserving existing XSysInfo and Kickstart assertions.
• Documents the corrected rule for hi-res early DDF in docs/internals/video.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/video/bitplane.rs	Removes the unconditional hi-res early-DDF snap and clarifies why window clipping/origin math is sufficient (fixes overscan window-left cases).
src/video/bitplane/tests.rs	Renames/updates the relevant test and adds an ECS extreme-overscan regression case asserting the correct origin (18).
docs/internals/video.md	Records the “hi-res early DDF is beam-anchored; window comparator hides pre-window fetch” rule and cites the overscan scenario (issue #186).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.