reMarkable 2 support (ARM32 + AppLoad/qtfb)

[mariomosca]

What

Makes riddle run on the reMarkable 2 (rM2, ARM32) under xovi + AppLoad in qtfb mode, alongside the existing Paper Pro takeover path. Everything is gated on the takeover flag / runtime backend, so Paper Pro behaviour is unchanged.

I ran it end-to-end on an rM2 (firmware 3.27.3.0): the page opens, pen strokes render, the ink fades, and the oracle (Claude via an OpenAI-compatible endpoint) replies in a flowing hand. 🖋️

Why these changes

Debugging the qtfb backend on-device surfaced a few rM2/AppLoad specifics:

• pen.rs — find_marker_device() only matched "marker" (Paper Pro). The rM2 pen enumerates as "Wacom I2C Digitizer", so I also match "wacom"/"digitizer".

• qtfb.rs — the AppLoad shim's ServerMessage puts the InitResponse union at offset 4 (shm_key @4, shm_size @8), not @8/@16, and publishes the framebuffer as a POSIX object opened with shm_open("/qtfb_<key>"). The old offsets read the size as the key and produced a nonexistent /qtfb_<size> name → ENOENT. user_input events are likewise @4/@8/@12/@16/@20. Verified on-device against the shim wire format (matches the canonical zqtfb client).

• main.rs — in qtfb mode, don't open the raw evdev pen: the shim intercepts the digitizer and delivers pen events over the QTFB socket, so the raw device yields nothing and the event loop was skipping the qtfb pen path whenever pen_dev.is_some(). Now raw evdev is opened only in takeover mode.

• main.rs — a corner-tap quit for qtfb/AppLoad (there's no touch device there for the 5-finger tap): a finger tap in the top-right corner exits cleanly.

• main.rs — coalesce qtfb flushes at 20 ms (~50 Hz) for a more continuous stroke over the IPC round-trip.

• oracle.rs — strengthen the persona so it always replies in the writer's language (it was defaulting to English for non-English input). Happy to split this into its own PR if you'd prefer to keep this one hardware-only.

Notes

• No changes to the takeover/quill path; the takeover feature builds as before.
• Cross-built for armv7-unknown-linux-gnueabihf (rM2 has GLIBC 2.39, binary needs ≤ 2.34).
• Glad to adjust naming, gate flags, or split commits however fits the project.

[serenasensini]

Hi @mariomosca! It seemed we had the same idea :)

Great work on this PR - I’ve been testing and extending the rm2 port on my side, and I’d be happy to contribute the additional changes if useful.

Compared to #19, my branch currently adds a few follow-up pieces that seem complementary:
• rm2 feature gating + rm2 screen geometry (fb.rs)
• 32-bit-safe evdev parsing (EV_SIZE/EV_OFF) in pen/touch paths
• rm2 pen mapping calibration (axis swap + Y inversion) and configurable pressure threshold (RIDDLE_PEN_MIN_PRESSURE)
• rm2 touch device support (pt_mt)
• quill 32-bit QImage constructor hook (needed to capture framebuffer pointers on rm2)
• build/bundle script parametrization for DEVICE=rm2 / SDK=...
• onboarding/docs for rm2 bring-up

I don’t want to step on ongoing work here, so I wanted to ask what you’d prefer:
• I can coordinate with this PR and split/cherry-pick only the missing parts, or
• I can open a separate follow-up PR scoped only to the additions above (referencing #19).

Happy to proceed in whichever way is easiest for review/merge.

You can find my fork and updated changes in here.