Fix: Correct timestamp units in doc comments

[bclarke123]

What

Corrects documentation that describes timestamps as Unix epoch seconds when the values are actually milliseconds: the v1 proto comments (Repository.created, Branch.created, thin-client commit timestamp), the generated C header, and the revision history --date help. Doc-only, no wire or on-disk values change.

Why

The implementation is uniformly milliseconds end to end util::time::timestamp(), the persisted revision-metadata TIMESTAMP key, the server's cache-age math, the CLI's from_timestamp_millis displays - so external consumers that trust the documented unit get values 1000× larger than promised (we hit this building a web frontend against the API and had to add a runtime unit-sniffing workaround).

The --date flag is the one place this becomes a functional bug: passing epoch seconds as documented compares against millisecond values, so the filter silently never matches.

Notes

• Audited all timestamp producers/consumers; the implementation never disagrees with itself - only the docs were wrong.
• Fragment last_access genuinely is seconds (as_secs()) and is left as-is.
• lore.h and prost output are regenerated, not hand-edited.