feat(input): bind Plan 9 arrow-key runes (fixes down-arrow in alacritty9)

[mparrett]

What

Movement keys bound only the ANSI arrow escapes (ESC[A/B/C/D). This adds Plan 9's kernel arrow runes (Kup U+F00E, Kdown U+F800, Kleft U+F011, Kright U+F012) alongside them, so the arrow keys drive movement on the Plan 9 terminals that deliver runes instead of escapes.

Why it's not cosmetic

Which form a terminal sends isn't uniform, and the gap lands on the main play environment:

• alacritty9 — where the game is actually played on 9front — sends ANSI for up/left/right, but the raw Kdown rune U+F800 for down. Measured on 9front amd64: without the U+F800 binding, up/left/right move and down does nothing.
• The bare text console sends all four keys as runes.
• rio's own windows swallow the arrows for scrollback, so they never reach the program — not something a key binding can address.

The down case is the one that bites: a silent dead key in the primary environment, easy to miss because hjkl and the numpad still move you.

Evidence

Probed on real hardware (9front amd64 + alacritty9) with a read-key codepoint dump, entering the alternate screen the way the game does:

↑  codes=(27 91 65)  ESC[A
↓  codes=(63488)     U+F800   ← Kdown rune, every press
←  codes=(27 91 68)  ESC[D
→  codes=(27 91 67)  ESC[C

The let-go side (read-key tokenizing these on Plan 9) is nooga/let-go#461; this is the xsofy-side binding that consumes them.

Test

input_test.lg covers the rune bindings resolving to the right movement actions.

Closes #157.

[mparrett]

Bare-console confirmation: all four runes on real hardware

The PR body measured only the alacritty9 down-arrow; the other three rune values came from keyboard.h. Ran the arrow probe on a bare 9front console (rio exited, raw /dev/cons) and all four match the bindings:

↑  codes=(61454)  U+F00E  Kup
↓  codes=(63488)  U+F800  Kdown
←  codes=(61457)  U+F011  Kleft
→  codes=(61458)  U+F012  Kright

(raw="" because they're Private-Use-Area codepoints with no glyph — the codepoints are the receipt.) Each arrived unbuffered, one rune per press, under consctl raw mode; q quit cleanly with the console back in cooked mode.

So the two device paths together cover every key this binds: the bare console sends all four as runes, and alacritty9 sends ANSI for up/left/right plus the Kdown rune for down.

[mparrett]

Just a hunch, this PR seems like it's trying to fix something that shouldn't be that way in the first place. The "raw Kdown rune U+F800 for down" part smells like a bug rather than something to accommodate here- Not proven, but need to take a deeper look.