UI: make Evidence Inspector & Workflow Dashboard legible to non-experts

[escapeboy]

Improves the two read-only --features serve web UIs so a non-expert can understand them (feedback: "I understand nothing about this interface"). Read-only, self-contained, XSS-safe — no behavior change.

Evidence Inspector (evidence_serve.rs)

• Lead paragraph defining what an evidence bundle IS; verification hint explaining what VALID/INVALID proves
• Plain-language hash legend; full-value-on-hover (title) for long hashes
• <caption> / <th scope> on tables, horizontal-scroll wrappers
• Helpful empty states (audit / outputs), active-nav aria-current, footer stating read-only/no-auth/127.0.0.1 posture, prefers-color-scheme dark mode

Workflow Dashboard (dashboard.rs)

• Shared header/nav with active cue; per-page explanatory copy
• Color-coded status badges (running/paused/completed/failed/pending)
• Semantic markup (<th scope>, <caption>, <main>, viewport meta), scrollable tables
• Footer clarifying it cannot mutate runs (CLI-only); dark mode

Constraints honored: all CSS inline (no CDN/network), read-only (no new mutating routes/forms), routes + JSON API shapes unchanged, every dynamic value still through esc(). Gated green locally: build + clippy -D warnings + fmt + tests (234). 2 files, +262/-74.

[github-actions]

Bench compare

Threshold for regression: ≥ 10% slower mean.

Benchmark	Mean change	99% CI
compile_crud_admin_template	-1.12%	[-2.39%, +0.15%]
compile_medium_program	+10.60%	[+7.11%, +14.36%]
compile_small_program	+2.41%	[-1.43%, +6.33%]
evidence_build_5_steps	-59.64%	[-74.38%, -28.34%]
evidence_build_empty	-17.33%	[-36.77%, +7.12%]
evidence_verify_10_steps	-33.58%	[-35.64%, -31.35%]
evidence_verify_5_steps	-28.93%	[-31.98%, -25.50%]
vm_call_dispatch_loop/iters=1000	-16.28%	[-20.18%, -12.22%]
vm_call_dispatch_loop/iters=10000	-12.84%	[-16.90%, -8.69%]
vm_pure_loop/iters=1000	-39.27%	[-40.99%, -37.33%]
vm_pure_loop/iters=10000	-40.88%	[-43.87%, -37.87%]
vm_pure_loop/iters=100000	-29.29%	[-32.62%, -25.55%]
vm_record_loop/iters=1000	-1.04%	[-2.71%, +0.35%]
vm_record_loop/iters=10000	-6.37%	[-9.74%, -3.29%]

⚠️ 1 benchmark(s) regressed past threshold:

• compile_medium_program