Problem: `parse.body()` scans tokens right-to-left and breaks on the
first non-metadata token. Forge refs like `gl:a/b#12` halted the scan,
preventing metadata tokens to their left (e.g. `due:tomorrow`) from
being parsed. Additionally, `diff.parse_buffer()` ignored
`metadata.priority` from `+!!` tokens and only used checkbox-derived
priority, and priority updates between two non-zero values were silently
skipped.
Solution: Recognize forge ref tokens via `forge.parse_ref()` during the
right-to-left scan and skip past them, re-appending them to the
description so `forge.find_refs()` still works. Prefer
`metadata.priority` over checkbox priority in `parse_buffer()`, and
simplify the priority update condition to catch all value changes.
c9790ed3bf
