fix(webui): match the TUI's per-character playbook run shimmer

[edwin-zvs]

What

The web Playbook editor now shimmers pending blocks with the same wave the TUI paints, instead of its own unrelated highlight band.

The TUI behavior I matched

crates/cli/src/ui.rs → apply_playbook_shimmer re-emits every character of a pending block with a brightness drawn from a travelling sine wave:

	TUI
temporal speed	PLAYBOOK_SHIMMER_SPEED = 4.2 rad/s
spatial frequency	PLAYBOOK_SHIMMER_DENSITY = 0.18 rad/char
brightness	smoothstep(0.5 + 0.5·sin(phase − gidx·D)), blended theme.muted → theme.text
continuity	one character index (gidx) advances across all pending lines, so the band runs continuously down the block
spans with a background	left intact, but still advance the wave

That works out to a 34.907ch wavelength (2π/0.18) travelling one wavelength per 1.496s (2π/4.2) — ~23.3 chars/sec, left to right and on down the block.

The web editor was doing something else entirely: a translucent --accent band swept across each pending row independently on a 1.6s cadence, restarting per line.

The change

crates/daemon/assets/index.html only (CSS + the existing playbookApplyShimmer):

• .playbook-line.is-running paints a 17-stop repeating gradient sampling that same smoothstep, sized 34.907ch, clipped to the glyphs with background-clip: text, and translated one wavelength per 1.496s.
• playbookApplyShimmer sets a per-line --pb-shim-lag (negative animation-delay) from the running character offset. The wave is periodic, so lagging a line by chars / 23.3 seconds reproduces the TUI's single gidx — the crest carries on across lines instead of restarting per row. It is anchored to the run's own start time, so a block that joins a run late stays in phase.
• Chips and inline code carry their own background — the TUI overlay skips those — so their colors are pinned back against the inherited transparent text fill. Fenced-code rows re-issue their block background as a border-box layer so background-clip: text can't clip it away.
• prefers-reduced-motion gets a static accent tint at full contrast instead.

Source of truth is unchanged: the daemon's run pending set (pending_block_refs), exactly as before. No local timers, no heuristics.

No per-frame JS and no idle repaint loop — it's a CSS animation that only exists while a line carries .is-running. With everything settled: 0 running lines, 0 leftover custom properties, document.getAnimations() reports 0 shimmer animations.

Known deviation

The TUI also sets BOLD when eased > 0.85. CSS can't vary font weight per character inside one text run, so the crest peaks at full --fg without the extra weight.

Evidence

Isolated daemon from this worktree (CONSTRUCT_ASSETS_DIR hot-reload), a real playbook run with 4 pending blocks declared over IPC, driven in Chrome over CDP.

TUI (the reference, unchanged):

Web, dark theme:

Web, light theme:

Two frames ~370ms apart, showing the crest travelling right and carrying across the two Notes: lines:

frame 1	frame 2

Also verified in the live page:

• Typing inside a shimmering line — inserted ZQX mid-line: text lands at the caret, caret offset advances correctly, line keeps .is-running and its animation; backspacing restores it.
• Selection stays legible (Chrome keeps painting the clipped gradient through the selection highlight).
• Settling stops promptly — settling 2 of 4 blocks over playbook.update cleared .is-running and --pb-shim-lag on exactly those lines and re-phased the rest; settling all cleared everything.
• prefers-reduced-motion: animation-name: none, no gradient, -webkit-text-fill-color back to rgb(236,236,241).

Testing

• cargo build — clean.
• cargo test --workspace --no-fail-fast — all green, including crates/e2e/tests/web_smoke.rs against a freshly built target/debug/construct (verified the new CSS is baked into the binary). One earlier run tripped pty_render::tests::many_small_events_stay_fast_as_history_grows, a load-sensitive perf test unrelated to this diff; it passes in isolation and in the full re-run.

Note for the reviewer

Another branch is concurrently editing this same file (C-space selection fix). This diff is confined to the shimmer CSS block and the body of playbookApplyShimmer.

The media above lives in a temporary .pr-media/ commit that is removed by the next commit on this branch; the links are pinned to 65ab37db5ed19b26aef683c2b333e2dfb0cb5a97 and stay resolvable via refs/pull.