fix(terminal,parser): honour SGR colon subparameters

[roramirez]

Problem

Any TUI that styles text with the extended CSI 4:<style> m form left the whole screen underlined for the rest of the session. Reproduced with Claude Code's session picker: after returning from the list, every line printed afterwards was underlined.

Root cause is in csi_dispatch, which flattened parameters with params.iter().map(|p| p[0]) and discarded everything after a colon:

• CSI 4:3 m (curly underline) → read as bare 4 → underline on (correct by accident)
• CSI 4:0 m (underline off) → also read as bare 4 → underline on again

There was no way to turn it back off, so the attribute leaked into every subsequent cell. The same flattening silently dropped the colon form of 38/48, so 38:2:r:g:b colours were ignored entirely.

Fix

csi_dispatch now collects Vec<&[u16]> — the full subparameter groups, with no per-parameter allocation — and passes them to handle_sgr alongside the existing flat view used by the legacy semicolon forms.

• 4 reads its style subparameter: 4:0 disables, 4:1..4:5 enable (we render a single underline style), a bare 4 still enables.
• The :3 in 4:3 no longer leaks through as SGR 3 (italic).
• New parse_color_from_subparams handles 38:5:n, 38:2:r:g:b and 38:2::r:g:b (colour-space id slot present), for both fg and bg.

Tests

Nine unit tests in parser_test.rs cover each form in isolation: underline on/off by subparameter, the mixed 1;4:0;3 case, the no-leak-into-italic case, and the colon forms of truecolor and 256-colour.

The anchor is the end-to-end scenario scenario_tui_curly_underline_does_not_leak_after_exit, which replays the real shape of the bug — enter the alternate screen, draw with 4:3/4:0, exit, keep printing — and asserts on the resulting cells, not on the parser flag.

Verified it actually catches the regression: reverting only parser.rs and keeping the test makes it fail on the "must not be underlined" assert; with the fix it passes.

cargo fmt --check clean, cargo clippy --locked -- -D warnings clean, cargo test 1512 passed / 0 failed.

Manual check

Inside mmterm, printf '\033[4:3mA\033[4:0mB\n' — A underlined, B not. Before this change both were underlined, along with everything after them.