Opt-in usage reports #176

We currently fly blind: no idea how many people use SpeakType, on which systems, with which models, or where it breaks. Every report is a bug that someone bothered to write up.

Opt-in usage reports would help, but only if they're done so plainly that nobody has to trust us.

How it has to work

• Off unless turned on. No pre-ticked box, no "improve SpeakType" dark pattern, no prompt on first launch.
• Never any content. No transcripts, no audio, no file names, no app names, no dictionary entries. Ever.
• Show what is sent, in the app. The exact payload, readable, on the same screen as the toggle.
• Documented in the repo, so it can be checked against the code.
• Off in one click, and the setting is respected immediately.

What would actually be useful

• System and version, so we know who a release reaches.
• Which model people choose, and whether the recommendation was taken.
• Whether dictation succeeded, failed, or found no speech, as counts.
• Crashes and errors by type, with no user text in them.
• Nothing that identifies a person or a machine beyond a random id that can be reset.

What it changes publicly: the README and roadmap currently say no telemetry, full stop. That has to become an accurate sentence instead, something like "nothing is collected unless you switch reports on, and here is exactly what they contain". The promise only holds if it stays true for people who never touch it.

Related: the optional account in #175.