Paint frq with jvui, and take the last Rust out of the window

glimmer-vidya painted through libvidya, which meant jolt-native's Cargo
workspace, which meant a build. glimmer-jvui paints through jvui — dvui's
shape written in jolt on SDL3, with no shared object of its own — so the
window is jolt and C and nothing else.

The alias moves with it: `vidya/` reads as a lie once nothing here is
vidya, so it is `gui/` in all four namespaces that call it. Every call site
is otherwise unchanged, which is what the backend being built to
glimmer-vidya's surface bought.

libvidya leaves the closure. libjolttui stays and is now the only object
out of that workspace frq loads at all — `just tui` and nothing else. What
arrives instead is SDL3, SDL3_ttf and SDL3_image, which jvui declares
itself and dlopens by soname. sdl3-image keeps its library in a separate
`lib` output; the default one holds only share/, which cost an afternoon
once and is written down in the flake so it does not cost a second.

TWO paths for one dependency: glimmer-jvui's deps.edn names jvui by
:local/root, a relative path that means nothing once nix has copied the
subtree out, so the -Sdeps has to name both.

Running it found the bug that reading it would not have. frq computes its
layout from `(long (second (screen-size)))` on the first tick, which runs
BEFORE the window opens — and screen-size answered nil there, which is a
NullPointerException in the caller rather than a small first frame. It
answers [0 0] now, the way window-width already did, and is right on the
next tick.

Verified: `just tui --headless` renders the sign-in screen, and the window
itself runs five minutes under SDL's dummy video driver with no exception.
Not verified: what it LOOKS like. There is no display here, so nobody has
seen frq painted by jvui — the tags are all present and the pixel tests
pass for the ones with a colour, but spacing, wrapping and the feel of it
want an eye.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>