Serve the compiled CSS instead of the committed main_pretty.css

[amjjbonvin]

Addresses backlog item 1, and fixes a live styling bug found along the way.

Problem

_includes/_head.html loaded assets/css/main_pretty.css — a committed,
hand-optimised artifact. Nothing in the repo regenerated it (the Gruntfile had
no CSS task, and that file is now gone), so it had drifted from the Sass sources.
Editing anything under _sass/ had no effect on the published site.

	Selectors
Compiled main.css (from _sass/)	1120
Shipped main_pretty.css	1115

In the sources but never shipped:

Selector	Status
.btn-assistant-professor, :hover	live bug — see below
#main article.fullpage, … img	dead — layout emits <body class="fullpage"> and <article class="people">, so it never matched
.clearfix	IE7 *zoom hack; .clearfix:after was shipped

Shipped but absent from the sources: .article-wrap p>a.prompt, li>a.prompt

The live bug

_layouts/people.html builds badge classes dynamically as btn-{{ stat }}, and
the people page emits btn-assistant-professor. The rule exists in
_sass/elements.scss but had never reached main_pretty.css:

$ curl -s https://www.bonvinlab.org/people/ | grep -o 'btn btn-assistant-professor[^>]*>[^<]*'
btn btn-assistant-professor">Assistant professor

$ curl -s https://www.bonvinlab.org/assets/css/main_pretty.css | grep -c btn-assistant-professor
0

That badge renders on the live site today with no role colour. This PR fixes it by
making the compiled stylesheet the one that ships.

Changes

• Port .article-wrap p>a.prompt / li>a.prompt into _sass/elements.scss — the
  only rule that existed solely in the generated file
• Point _includes/_head.html at assets/css/main.css
• Delete assets/css/main_pretty.css
• Set sourcemap: never, so the 153 KB main.css.map is no longer deployed now
  that main.css is public
• Document the styling workflow in README.md

Verification

• Nothing is lost. The old shipped stylesheet is a strict subset of the
  compiled one: 0 selectors lost, 5 gained, and of those five only
  .btn-assistant-professor matches any markup.
• Across the whole site, the only HTML difference against the live pages is
  the stylesheet href.
• Served CSS: 56,479 → 60,454 bytes uncompressed. That is the cost of dropping the
  extra selector-merging pass that the untracked optimiser used to apply; it
  compresses away substantially over the wire.

Role badge styling

_layouts/people.html builds badge classes dynamically as btn-{{ stat }}. Three
of the values it emits had no role colour on the live site. All three are now fixed:

Badge	Problem	Fix
.btn-assistant-professor	rule existed in _sass/, never reached main_pretty.css	fixed by serving the compiled CSS
.btn-visiting-phd	no rule anywhere, in either stylesheet	grouped with .btn-phd ($warning)
.btn-visiting-postdoc	no rule anywhere, in either stylesheet	grouped with .btn-postdoc ($info)

The two visiting roles are grouped with their base role rather than duplicated or
@extend-ed, so they stay in step and each compiles to a single rule.

Verified: every badge class the people page emits now has a matching rule.

[amjjbonvin]

Force-pushed to drop b3482cd (the _sass/forms.scss cleanup), which has been split out into #792 on review feedback. It was unrelated to the CSS pipeline fix here and undocumented in this description — a fair catch. It cherry-picked onto master cleanly, confirming independence.

This PR now contains three commits, all within its stated scope:

1. ddada5d — serve the compiled main.css instead of the committed main_pretty.css (the drift fix, which also fixes the unstyled .btn-assistant-professor badge)
2. 08b7653 — style .btn-visiting-phd
3. aada3df — style .btn-visiting-postdoc

The two badge commits are described in the "Role badge styling" section above. They belong here rather than in a separate PR because they are only observable once this PR lands — on master the served stylesheet is the committed artifact, so a new rule in _sass/ has no effect until the pipeline change is in.

[amjjbonvin]

Copilot review:
This is still a good PR overall, and the main fix is the right one: the site should ship the generated CSS from _sass/ instead of a stale committed artifact.