fix(cli): messages edit should read '-' from stdin like messages send

[MrSucik]

messages send documents --content - as "read from stdin" and implements it via
read_or_stdin. messages edit documents nothing and never calls the helper, so
--content - is stored as a literal -.

That is worse than a missing feature: the flag is the natural way to pass a long or
shell-hostile body, it silently succeeds, and it overwrites the message being corrected.
We hit this on a production channel — an edit meant to fix a meeting summary replaced it
with a single bullet.

This routes edit's content through the same helper, documents the flag, and adds an
example to --help. Also adds a read_or_from seam so the sentinel behaviour is unit
testable without touching the process's stdin. read_or_stdin keeps its existing
signature and delegates to the seam, so no caller changes.

Testing

Toolchain 1.95.0 from rust-toolchain.toml. just is not installed on this machine, so
the Justfile's own commands were run directly; cargo-nextest is absent, so cargo test
was used where test-unit prefers nextest.

• cargo fmt --all — clean, no changes outside the three edited files.
• cargo clippy -p buzz-cli --all-targets -- -D warnings — passes. (Scoped to buzz-cli
  rather than the Justfile's --workspace; the change touches only this crate.)
• cargo test -p buzz-cli — 474 passed, 0 failed, including the new
  validate::tests::read_or_from_dash_is_a_sentinel_not_literal_content.

Manual end-to-end proof against a local isolated relay
(./scripts/start-isolated-test-relay.sh, no credentials, no shared or production relay),
A/B on the same relay with the same command:

printf 'x\ny\n' | buzz messages edit --event <ID> --content -

Reading the resulting kind-40003 edit event back out of the relay's database:

CLI build	stored content
main (unpatched)	-
this branch	x\ny\n

As the author notes, the unit test alone would not have caught the original bug — it was a
missing call in edit, not a defect in the helper. The manual A/B above is what actually
covers the regression; the crate has no CLI-surface test harness to encode it in.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[github-actions]

🔐 Codex Security Review

Status: review required for the current range.

The current range is a9dd4d3bd771b47f0efb4e7f6f7329a5921372b5...20a00df0c056b432733e8c68438b0d5bb7078476.
A new review must complete for this exact range. When manual authorization
is required, a Block organization member must comment exactly
@buzz-security-review 20a00df0c056b432733e8c68438b0d5bb7078476 to authorize a new review.
Any previous review applies only to its recorded range.