H5P bridge polish (from the live field test)

[aharlap]

Two real findings from the 2026-07-08 real-browser H5P field test (examples/h5p-demo + pnpm field), both non-blocking — the bridge works end to end:

• Learner timeline shows the raw subContentId, not the question title. The answered event's H5P question title ("Which fraction is largest?") IS captured — stored as the /q/<id> child activity's name — but the learner timeline's Detail column renders the raw id (abc-123). timelineDetail parses the id from the IRI; it should look up the stored child-activity name (like the funnel's stepLabels does) and fall back to the id. Same treatment would improve the funnel/answered rows.
• Subscription race on the content-load attempted event. H5P fires attempted when content instantiates; Proof's bridge polls every 250ms to find H5P.externalDispatcher and subscribe. On a fast load the first attempted can fire before the subscription exists, so the initial 'Started' is missed (later interactions land fine). Options: shorten the initial poll cadence, subscribe eagerly if H5P is already present at init, or document that content-load starts may be missed. Low impact (answers/completions unaffected); worth a note in embed.md at minimum.

🤖 Generated with Claude Code