Honor --canonical-url on every network that has one

[ralyodio]

--canonical-url reached dev.to and nowhere else. So --to htmlblog,devto
published the same article twice with no statement about which was the
original, and a Hashnode or Ghost copy had no way to make one at all.

What each network actually calls it

This failed quietly rather than loudly because the field name is different
every time:

network	field
dev.to	canonical_url (already worked)
Hashnode	originalArticleURL
Ghost	canonical_url
Tumblr	source_url — the attribution link it has instead of a canonical
gitblog	canonical: frontmatter key, for the site template to render
htmlblog	a real <link rel="canonical"> in the head

WordPress and Micro.blog are deliberately excluded. WordPress core has no
canonical field — it belongs to an SEO plugin's post meta — and Micropub
defines no canonical property. Wiring something plausible there would ship a
flag that looks configured and does nothing.

htmlblog points at itself

The original should confirm what the copies claim, so a page gets a
self-referential canonical from the siteUrl the account logged in with.
--canonical-url overrides it for an article first published elsewhere.

A canonical that does not byte-match the real address points at a different
page, so postUrl() is now the single place that joins siteUrl to a file
name, shared by the canonical, the URL myna reports and the timeline. It trims
trailing slashes, which the old inline concatenation did not — that was a
latent double-slash bug in the reported URL.

The blog-post seam

myna post --to htmlblog does not write the page itself when
profullstack/cli-tools' blog-post is on PATH, and on the dev box it is. So a
tag added only to myna's native writer would never have reached the real blog.
blog-post applies its own siteUrl (profullstack/cli-tools#55) and myna
forwards --canonical only when one was explicitly given. That flag needs
cli-tools 0.28.0.

Verified

• 242 tests pass, 12 skipped, 11 new
• The API adapters are driven against a stubbed fetch and the request body read
  back, since the assertion worth making is the field name on the wire. One
  test loops every network claiming support and fails if the URL is missing
  from the request, so adding an adapter and forgetting to wire it breaks.
• Also covered: HTML escaping of the URL, the no-siteUrl case emitting no tag,
  frontmatter quoting, and trailing-slash joining
• tsc --noEmit clean

Version bumped to 0.12.0 across the packages that track core. The plugin
packages sit at 0.8.4 on main already; that drift is pre-existing and untouched.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GNpSGiTnfodDyCQ3MU6Nsk