v0.1.1

The two issues from the first use in another repository, plus housekeeping after the
release.

• Record the v0.1.0 DOIs (concept 10.5281/zenodo.22649033, version 10.5281/zenodo.22649034) and add the PyPI, CI and DOI badges.
• CI passes CODECOV_TOKEN and fails the upload loudly: five runs had uploaded nothing while green.
• Coverage measures the CLI and every subprocess (the parallel + patch=subprocess modes);
  in-process CLI tests and the remaining core branches take it from 72 % to 100 %.
• A relative --python (or PYTHON) passed the existence check, then crashed with a
  traceback from the sandbox (#1): the check ran in the caller's cwd, pytest in the
  sandbox. The interpreter is now made absolute once, before the run: --python relative
  to the caller's cwd, a declaration's PYTHON relative to the root like ROOT and
  PATHS (a bare name is still looked up on PATH).
• test_run_verbose_only_keep_and_stop no longer leaves a sandbox in $TMPDIR on every
  suite run (14 per self-check), and now asserts the kept directory exists (#2).
• Three new self-contracts pin the fixes (python-unresolved, keep-ignored,
  declared-python-cwd-anchored); the self-check reads 16 OK.

v0.1.0

First release. Mutation with fires / may_fire / invisible / count; the sandbox
copy via git ls-files; verdicts OK, DECORATION, OVERREACH, VISIBLE, NOT_APPLIED, ERROR;
a red baseline aborts; mutgate run and mutgate list, with --markdown for build
records. Extracted from three hand-rolled mutation harnesses written in one week across
the Morphospace and abiogenesis repositories, after the third one found a duplicated
convention by firing on the wrong kind of assertion.

Fixed before release, on a second session's review of the first commit

• A symlinked TMPDIR crashed every mutation ("escapes the sandbox"): the sandbox
  directory is now resolved before the containment check. macOS keeps /var behind a
  symlink, so this was every macOS user.
• A suppressed pytest summary was a silent false pass. pytest exit 1 with no
  failure readable is now ERROR, never OK; the same at baseline.
• fires="Name" (a bare string) was a silent false pass: it became one-letter
  fragments matching every node id. Strings and empty fragments are refused.
• A typo in --only, or an emptied MUTATIONS, exited 0. Both are usage errors (2).
• A project whose config stops at the first failure (-x, --maxfail=N) truncated the
  fired set with a consistent-looking exit, so an OVERREACH read OK. mutgate now owns
  --maxfail=0 after every user argument, and pytest's own "stopping after N failures"
  line (a conftest can still force it) makes the run ERROR. Second review round.
• An interrupted run with a partial fired set read as a verdict (a conftest setting
  session.shouldstop after the first failure: exit 2, one failure parsed). Only exit codes
  0 and 1 now count as a completed run; anything else is ERROR whatever parsed. Third
  review round.
• Node ids containing " - " inside a parametrisation are no longer truncated; --timeout
  makes a hanging suite an ERROR; a missing --python is a usage error; ROOT is
  relative to the declaration file; the git ls-files sandbox and real PYTHONPATH