test: analyse need for golden-file assertions on generator/parser HTML output

[harrydayexe]

Is your feature request related to a problem? Please describe.
The product's primary output — generated HTML — is asserted with weak substring
checks at the generator and parser layers (e.g. contains "<h1",
"Test Post", "<strong>valid</strong>"), and there are no .golden expected-
output files. A regression that mangled surrounding markup, dropped an
attribute, reordered sections, or corrupted escaping while still emitting the
asserted substring would pass unnoticed.

Note this is scoped to the generator/parser rendering layers only — the
outputter already does byte-for-byte content comparison
(directoryWriter_test.go), so the write layer is not in question.

Describe the solution you'd like
An analysis (not necessarily implementation in this issue) of whether golden-file
/ byte-exact assertions should be introduced for a few representative renders.
The analysis should produce a recommendation covering:

• Which documents are worth snapshotting (e.g. a post with code + footnotes, the
  index, a tag page) vs. keeping substring checks.
• A normalisation strategy for volatile fields (e.g. Year/date) and volatile
  markup (goldmark version bumps, syntax-highlight spans).
• Whether to adopt a -update flag convention for regenerating goldens.

Describe alternatives you've considered

• Keep substring assertions as-is (cheapest, but weak on the Add parser and logging #1 named failure
  mode: incorrect HTML).
• Golden-everything (rejected pre-emptively — churny, breaks on every goldmark
  bump).

Additional context
Raised by the testing audit (golden-file finding, Medium). This issue only asks
for the analysis/decision; any implementation can be split out once the scope is
agreed.