#108 parse_ts_secs panics on multibyte timestamps

Problem

parse_ts_secs slices the offset by byte index without checking a char boundary. A hostile but valid UTF-8 .forge timestamp can panic the whole build. src/build.rs:599.

Verified reproduction

.forge/releases/v1.md:

created: "2026-01-01 12:34:56+€!"

Build:

cargo run -- build /tmp/rev5-ts -o /tmp/rev5-ts-out
exit=101
thread 'main' panicked at src/build.rs:599:27:
end byte index 11 is not a char boundary; it is inside '€' (bytes 9..12 of string)

created/updated from issues, PRs, comments, reviews, releases, and CI runs all go through time_ago -> parse_ts_secs. This is the same class as #99 (multibyte truncation) but in a different parser.

Fix

Require the time/offset suffix to be ASCII before slicing, or guard each str slice with is_char_boundary and return None. Add a multibyte timestamp regression test using the repro above.