v0.6.2

Replies from a fallback account return without a pause

When a request was served by a fallback account, the plugin recorded a "last used" timestamp before handing the reply back — and on a machine running several sessions at once, that write had to queue behind every other session's write to the same file.

The reply was already sitting there while you waited. Measured on a busy machine: about 3.6 seconds typically, up to 15 seconds at worst, on every fallback-served request.

The timestamp is now recorded in the background. Nothing reads it to make a decision, so it was never worth waiting for.

Clearer account-store timeouts

If a write to the account store does time out, the message now reports how many attempts it made and how far apart they were. Widely spaced attempts mean the machine was too busy to run the retries, which is a different problem from the file genuinely being contended — and the two need opposite fixes.

The previous wording asserted that another writer had held the file for the whole wait, which measurement showed is often untrue.