chore(release): fold the post-rc.2 work into the 0.5.0 section

[donislawdev]

Closes the changelogs for 0.5.0. Four pull requests landed after v0.5.0-rc.2
was cut (#141 to #144), one of them the "Internet only" switch, and every note
they carried had collected under [Unreleased]. This moves them into the dated
section, so the tag ships what the release actually contains rather than what
the release candidate did.

The tag check cannot catch this on its own: an rc and its release share a
version by design, so v0.5.0-rc.2 and v0.5.0 both match 0.5.0.

What changed

• CHANGELOG.md: the moved entries join the Added, Changed and Fixed
  blocks [0.5.0] already has, rather than opening a second set of headings
  that the duplicate-heading guard would reject. The Added ones lead their
  block, because the new switch is the largest user-visible change in the
  section.
• The summary gains a third paragraph naming that switch, including the one
  thing worth knowing before ticking it: the router is on the local network too,
  so name lookups can stop with it.
• The section carries the release day, not the day the release candidate closed
  it.
• [Unreleased] stays as a bare heading, which is the shape the release
  workflow checks for at tag time.
• The internal changelog is closed the same way. It is not tracked here, so this
  pull request cannot show that half.

No code changed.

Verification

• tests/test_version_and_release.py - 25 passed.
• python tools/release_notes.py 0.5.0 extracts the section cleanly, 26 940
  characters.
• Neither file contains an em or en dash, and both stay pure LF.
• The internal pre-release rig sweep ran in full against this tree: every rig
  measured, nothing skipped and nothing refused. Six reported movement and all
  six were traced to the measurement rather than to the product - a peer that
  was not listening, a sentinel value being read as "no change", machine load
  during the run, a new counter belonging to the new switch, a random count
  inside a verdict that held, and one metric whose run-to-run spread is wider
  than the movement it reported, shown by running the same measurement against
  the older code interleaved with this one.

The full test suite runs on CI rather than locally.