Fix: check the system clipboard off the UI thread

[omellado-gh]

What and why

The app checks the system clipboard every 0.25 s to decide whether the Paste menu
items are enabled. On X11, arboard's clipboard read waits up to 4 s for a
SelectionNotify; when the clipboard owner never answers (a hung or foreign app,
a clipboard manager, XWayland), that call ran on the UI thread and froze the whole
frame loop — the window stopped drawing and could not be closed.

Reproduced with the real failure: seed the clipboard and stop its owner
(printf x | xclip -selection clipboard; kill -STOP $(pgrep -x xclip)), then open
a document. The app became unresponsive and a control-channel ui.inspect timed out
while the main thread sat in arboard's wait loop.

This change moves the check off the UI thread: a clipboard_probe worker owns a
second clipboard handle, publishes whether Paste has something to take
(has(PASTE_ORDER)) over a channel, and wakes the UI only when the verdict changes.
logic_frame reads the latest verdict with try_recv and never blocks. Without a
factory (the web, and hosts that don't install one) the check stays in line,
throttled, exactly as before. The factory is installed on Linux only for now;
macOS and Windows keep the in-line check, so their behaviour is unchanged.

How it was tested

• New tests in crates/ui-egui/src/tests_sysclip_probe.rs (5): a fake clipboard
  whose has() blocks proves the headless UI frame does not wait on it (measured
  elapsed time) and that the check runs on the vectorcraft-clip probe thread; the
  probe's verdict enables the Paste menu; a probe that reports nothing leaves it
  disabled; dropping the app stops the probe; and the in-line path (no factory) still
  works. cargo test -p vectorcraft-ui-egui --lib → 717 passed.
• End to end with the real X11 hang: with the clipboard owner stopped with SIGSTOP
  and a document open, the app stays responsive (ui.inspect answers in ~0 ms, main
  thread in its normal epoll wait), where before it froze.
• cargo check -p vectorcraft, cargo fmt --all --check, and clippy on the touched
  crates.

Note: locally, workspace clippy / cargo xtask ci fails on a pre-existing lint in
crates/eps/src/import/data.rs:746 (chunks_exact_to_as_chunks, new in clippy 1.99)
that this PR does not touch; the touched crates are clippy-clean with that lint
allowed.

Checklist

• cargo xtask ci passes (fmt, clippy, tests, assets, brands, layers, wasm).
• No panics in shipped code: no unwrap(), expect(), panic!, unreachable!, todo! or
  unimplemented! outside tests, and no unchecked indexing on data from files, parameters or
  messages. Errors are returned as Result and shown to the user
  (why and how).
• Clean-room: nothing copied from Adobe products or GPL/AGPL code (AGENTS.md).
• Every new asset has a row in ASSETS.md.

[bflatastic]

Thanks a lot for this, @omellado-gh, and welcome! A great first PR: you tracked the freeze down to arboard's X11 read on the UI thread, reproduced it with a stopped xclip, and wrote tests that catch it first. Because this PR comes from your fork's main branch, we landed it from our own branch, with your commit (squashed into one, authored by you) plus a follow-up on top, instead of force-pushing over your main:

• the background check only reads the clipboard while Paste depends on it (a document is open and nothing has been copied in the app), and wakes the UI only when the answer changes;
• Windows, macOS and the web keep their in-line check exactly as before, including its short-circuit;
• if the worker thread dies, the check moves back to the UI thread, and quitting waits a moment for the worker to drop its clipboard handle, so X11 still hands what you copied to the clipboard manager;
• we trimmed the tests to three that share the existing clipboard test helpers, and noted the change in ROADMAP.md and docs/development.md.

If you'd like to chat about Linux and the rest of VectorCraft, come join us on Discord: https://discord.gg/artcraft

[bflatastic]

Landed via #534 (your commit, authored by you, plus our review commit). Thanks again!