fix: maintain RFT on plain text

[rgdevment]

What does this PR do?

Two independent changes on this branch:

1. Formatting is no longer discarded on a plain re-copy (#56)

Copying text that is already in the history again, this time without styles, no
longer clears the rtf/html stored for that item. Stored styles are replaced
only when the incoming copy brings its own — and when it does, both keys are
replaced together, so a copy carrying only HTML can never leave the previous RTF
behind. An empty payload counts as "no styles" and clears nothing.

This reverses the drop introduced in #53. No new setting: it is the behavior for
everyone, on all three platforms.

2. Release hardening

Build provenance attestations for every published artifact (Windows installer
and MSIX, macOS DMG, Linux AppImage/deb/rpm/tar.gz), plus a update-scoop-bucket
job that publishes the manifest to rgdevment/scoop-bucket on tag, handling
stable and beta channels separately.

Why?

Rich text contains the plain text, not the other way around. Paste as plain
text already serves the unstyled version at paste time without touching what is
stored, so dropping the formatting layer destroyed the superset to keep the
subset — and left that very button with nothing to do, since a normal paste
already produced plain text.

The original reasoning behind #53 was fidelity: the card should mirror your last
copy. The counter-argument is asymmetry — losing formatting is irreversible,
while unwanted formatting is one click away. On top of that, re-copying as plain
is almost never an instruction to forget the formatting; it usually just means
the text was copied from somewhere else.

Deduplication stays exact (content, type). No fuzzy or "similar" matching, and
no separate rich/plain entries.

Closes #56.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!