feat: a folder entry can exclude a checkout by name

[tschm]

Follow-on to #32. A folder entry is only usable for a directory that is
entirely yours, and the one in front of me was not: ~/repos/jebel-quant
held 26 of my own repos and a clone of upstream numpy. Listing 26 repos to
leave out the 27th is exactly the terseness folder: was meant to remove.

repos:
  - folder: ~/repos/jebel-quant
    exclude: [numpy]      # a clone of upstream numpy, not ours

A single name needs no brackets.

Excludes name directories

That is the one thing about a child that is knowable without reading its git
config, and the thing you see when you list the folder. The two really do come
apart — ~/repos/tschm/rhiza_projects/funnel is VanekPetr/funnel — so which
of the two is meant had to be settled rather than guessed. An owner/name
written there is refused rather than quietly matching nothing.

A spent exclude warns, and does not refuse

Every other mistake in repos.yml is fatal. This one is not, and the asymmetry
is the point of the change:

• The harm those refusals guard against is a repo silently missing from the
  board.
• An exclude exists to keep a repo off the board. A folder with no such
  checkout in it already satisfies that request — the outcome is the one the
  file asked for, and only the line is spent. That is what deleting a clone you
  never wanted looks like: a normal thing to do, which must not take the board
  down.
• It is also what a misspelt exclude looks like, and that one does put an
  unwanted repo on the board — so it is logged. But it lands as a row you can
  see next to the warning, not as the absent row you notice weeks later.

exclude on an entry that is not a folder is refused: one repo has nothing
to exclude from, and reading the key as decoration would leave you believing a
repo was off the board while it sat on it.

The first draft of this refused a spent exclude. numpy was then deleted from
that folder mid-change, and the refusal would have stopped the collector over a
directory whose removal was the whole point — hence the rule above.

Verification

• 447 tests pass; repos.py at 100% statement coverage; ruff check and format
  clean.
• Against the real fleet: repos.yml went from 27 entries to 5 (three folders
  and two paths), and the fleet went 27 → 38 repos with nothing dropped —
  10 more Jebel-Quant/*, tschm/cvxball, and tschm/pyhrp gaining the
  checkout it turned out to have all along.

🤖 Generated with Claude Code