Tracebacks where the exit-status table promises a diagnostic

[mairas]

From the review on PR #1: #1 (comment)

check-anchors reserves exit 2 for "the check could not run" and PR #1 extended that to translation-status --check, check-typography and check-glossary. The rest of the package has not caught up, and the failure mode is always the same: a raw traceback at exit 1, which is the status a workflow reads as "the check found problems".

• A missing or unreadable mkdocs.yml gives FileNotFoundError and exit 1 in every command that reads it, while check_anchors.configured_base() silently returns / for the same condition — in the function whose own docstring explains that a wrong base makes the checker report a pass it did not earn.
• blob_hash uses check=True, so any git failure surfaces as CalledProcessError. A *.md directory under a locale reproduces it (exit 128, uncaught).
• git is an undeclared runtime dependency; its absence gives FileNotFoundError from subprocess._execute_child. (Running outside a work tree is fine — hash-object needs no repository.) PR Package the documentation checkers and add a blocking translation gate #1 documented the requirement in the README; the diagnostic is still a traceback.
• configured_languages() calls next() with no default, raising StopIteration when no locale is marked default.
• stamp-translation writes files one at a time and raises SystemExit mid-batch on a bad path, leaving earlier files written and later ones not.
• No subprocess.run call passes a timeout=, so a hung git hangs the job.

Fix: one config-loading path that turns a missing or malformed mkdocs.yml into a one-line message at exit 2, and the same treatment for a missing git. Validate every path in stamp-translation before writing any of them.

The exit-status table is now in the README; the code should match it everywhere.