Add opt-in PromptSyntax deployment study analytics

[pathscale]

Summary

• add an off-by-default Research setting for a bounded PromptSyntax deployment study
• persist content-free WorkTable events for turns, PS operations, receipts, and comparable manual actions
• preserve authored-input metadata through compiled prompts, attachments, queued sends, and live follow-ups
• export user-initiated de-identified JSONL with per-export pseudonyms and support confirmed local deletion
• harden the event contract with fixed detail keys, typed failure categories, opaque identifiers, protocol and parser versions, and schema fingerprint coverage

Privacy boundary

The study table does not accept prompt text, agent prose, task titles, paths, URLs, tool calls, tool output, or attachment contents. Nothing is uploaded. Collection begins only after explicit opt-in and stops immediately when disabled.

Verification

• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace
• bun run lint
• bun run typecheck
• bun run test:run
• bun run build

This PR is intentionally stacked on #79 because it instruments the PromptSyntax authoring surface introduced there.