Reject invalid raw HTML tags, attributes and email autolinks

[bobzhang]

Split out of #152.

Bugs: invalid raw HTML and email autolinks were accepted.

Repro	Was	Rule
<33>, < a>, </1a>	raw HTML	tag names start with an ASCII letter
<b 1=c>	raw HTML	attribute names start with a letter, _ or :
<a b=>c>, <a b=`c>	raw HTML	unquoted attribute values are non-empty and can't start with "'=<>`
<foo\+@bar.example.com>	email autolink	the atext set had \ where ' belongs

Fix: tag_name checks the first character, attribute_name no longer accepts a digit first, attribute_value rejects an empty unquoted value, and char_is_atext_plus_dot has ' instead of \.

Spec examples fixed: 606, 618. Example 621 also needs #156 (the last line of the example is dropped without it).

Tests: src/cmark_html/invalid_raw_html_test.mbt. Updated raw_html_test.mbt (</1div>, <.div>) and raw_html_wbtest.mbt (tag_name on 1div), which encoded the old behaviour.

Depends on: nothing, independent of the other PRs.

[bobzhang]

Automated review: Codex CLI (codex exec, model reasoning effort high, read-only sandbox), 1 round, on git diff against the base branch.

Verdict: signed off, no changes requested.

I sign off on the code review. No concrete correctness, regression, or code-quality problems found.

The changes match CommonMark 0.31.2’s raw HTML grammar and email autolink rules. Tests cover all four fixes with valid counterparts. No unnecessary helpers were introduced.

Codex could not run moon test in its read-only sandbox. The tests were run separately on native, js and wasm-gc and pass.