Async enrollment FFI: completion-callback variants of enroll request/claim

[adamski]

Problem

replicant_enroll_request / replicant_enroll_claim are synchronous blocking calls (connect timeout ~10s, request timeout ~30s). Consumers calling from a UI thread stall the UI for the duration; today every consumer must spawn its own thread around a library primitive.

Enrollment is a replicant-protocol operation (endpoints, token exchange, credential format are all the library's contract), so the concurrency belongs in the library — consumers should only have to marshal the completion to their own thread.

Design (agreed 2026-07-14)

• Add additive (non-breaking) async variants alongside the sync ones:
  • replicant_enroll_request_async(base_url, email, cb, ctx)
  • replicant_enroll_claim_async(base_url, email, token, cb, ctx)
• Internally reuse the existing dedicated-thread + runtime-guard machinery the sync paths already use; invoke the completion callback instead of returning.
• Threading contract (document loudly): unlike the Client event callbacks (which fire on the consumer-pumped event thread), these fire on a lib-owned worker thread — necessarily, since enrollment runs before any Client/event pump exists. Consumers marshal to their UI thread themselves.
• Callback carries the same result surface as the sync versions (status/error + credentials out for claim); ctx lifetime rules documented (must outlive the call; no cancellation in v1 — the existing timeouts bound the wait).
• Update include/replicant.hpp and the JUCE wrapper; regenerate replicant.h with the pinned cbindgen.

Consumer side

entonal-common TonalDB will wrap these into std::function<void(Result)> completions marshalled onto its message thread (tracked in the Entonal Studio Linear project). The BLOCKING doc comments on TonalDB's enrollment API mark the current behaviour.

https://claude.ai/code/session_01GwMbERgLK4KAuLrz4b4Liw