fix(text): colour alpha, rich_text whitespace and baseline, gradient_text angle and stops

[LeadcodeDev]

Closes #366. Closes #367. Closes #368. Refs #388.

⚠️ One behaviour change to review before merging

#368 changes what an existing gradient_text angle renders as. The angle was applied as (cos θ, sin θ); it now follows the CSS convention (sin θ, −cos θ), so 0° is to top and 90° is to right. The default stays 90, and now genuinely renders left-to-right.

A scenario that set an explicit angle to compensate for the old axis will render differently. That is the fix, not a side effect — but it is worth a look at any file that tuned an angle by eye.

#366 — text ignored the alpha of style.color

paint.set_alpha_f(1.0) immediately after paint_from_hex(color) threw the parsed alpha away.

style.color's alpha channel (0x12 = 18/255, about 7%) must survive to the
painted pixels instead of being forced fully opaque; got max alpha 255

hex-colors.md already documented #RRGGBBAA generically — that claim is now true for text too.

#367 — rich_text: two independent bugs

Leading spaces. Tokenization always used split_whitespace(), collapsing runs. A white-space: pre|nowrap branch now keeps each span verbatim and disables wrapping, mirroring what text already does for those values.

white-space: pre must keep the leading spaces verbatim
  left:  "AB"
  right: "    AB"

The 9 px offset. baseline_offset was (line_height + ascent) / 2, missing the descent; text uses (line_height + ascent − descent) / 2.

rich_text's baseline must be computed as (line_height + ascent - descent) / 2
… got top 22, expected 13

A 9 px gap — exactly the issue's own measurement.

#368 — gradient_text: angle, line length, and stops

Beyond the angle convention above, the gradient line length was the box diagonal regardless of angle. It is now the CSS box projection (|w·sinθ| + |h·cosθ|) / 2, so the first and last colour land on the first and last glyph edge at any angle. The pre-fix symptom is the issue's own: the left glyph edge read (152, 0, 103) — muddy purple — instead of the first colour.

New optional stops (color + position), falling back to even spacing when omitted. Pinned as backward-compatible: even=(50, 205, 0), skewed=(50, 205, 0) was the pre-fix failure, identical colours proving stops was ignored.

Verification

Seven tests, each proven red before the fix by hand-editing the code back rather than by git stash (worktree safety). cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace (1601) all clean. Comment count still 0 — the only /// added are on GradientTextStop and the new stops field, both JsonSchema-derived.

Also found

The gradient_text early-return guard would have skipped rendering for a scenario using stops with an explicit colors: []. Widened, since it is the same code path.

Written comment-free, per the codebase-wide rule from #345.