Add opt-in parallel testing; fix registry and fixture isolation

[tony]

Summary

• Add opt-in parallel test execution via pytest-xdist (new dev dependency) with a just test-parallel recipe (uv run py.test -n auto). The default uv run pytest is unchanged and still runs serially — parallelism is purely opt-in.
• Fix VCSRegistry sharing parser state across instances. Its parser map was a class variable, so constructing any second registry mutated the shared registry singleton (replacing its git/hg/svn parsers). Each registry now owns its own parser map.
• Fix the git_repo / hg_repo pytest fixtures leaking state between tests. The first consumer of either fixture received a live handle to the session-cached master checkout, so mutating that checkout (adding a remote, switching branches) polluted the cache for every later test. The master copy is now a pristine, read-only cache and every consumer — including the first — gets its own copy.
• Net effect: the suite is now order-independent and safe to run under xdist. The two isolation fixes also matter to downstream consumers of VCSRegistry and the pytest fixtures (e.g. vcspull), independent of parallelism.

Changes by area

Parallel testing (opt-in)

• pyproject.toml: add pytest-xdist to the dev and testing dependency groups.
• justfile: add a test-parallel recipe — uv run py.test -n auto.
• .gitignore: ignore the .pytest-optimizer/ profiling-state directory.

Test-isolation fixes

• src/libvcs/url/registry.py: initialize parser_map as an instance attribute in __init__ instead of a ClassVar, so each VCSRegistry is independent and the module-level registry is never mutated by constructing another.
• src/libvcs/pytest_plugin.py: build the git_repo / hg_repo master checkout once as a pristine read-only cache, then always return an isolated copytree of it — the first consumer no longer receives a live handle to the cache.
• src/libvcs/url/base.py: unregister the demo rule at the end of the RuleMap.register doctest, matching the git/hg/svn examples, so it doesn't leak into the global GitURL parser.

Design decisions

• Parallelism is opt-in, not the default. The default uv run pytest stays serial and stable. The worker count is left to the operator (-n auto) rather than hardcoded in addopts, because the optimal count is machine-specific (high-core machines oversubscribe a subprocess-bound suite).
• Fix isolation at the source, not by constraining test order. Rather than pin test order or use xdist --dist grouping to paper over the coupling, the registry and fixtures were made genuinely isolated. This is the more robust fix and it also corrects real bugs for downstream callers.

Before / after — VCSRegistry

Before, a second registry clobbered the global one's parsers:

from libvcs.url.registry import registry, VCSRegistry
VCSRegistry({"git": MyGitURLParser, ...})
registry.match("git+https://github.com/o/r")  # now matched by MyGitURLParser

After, each registry is independent; the global registry is untouched.

Verification

The suite is order-independent (run shuffled across several seeds):

$ uv run --with pytest-randomly pytest -p randomly -p no:cacheprovider

Parallel execution is green:

$ uv run pytest -n auto

Test plan

• uv run ruff format . and uv run ruff check . — clean
• uv run mypy — clean (strict, covers src and tests)
• uv run pytest — default serial suite green and unchanged
• uv run pytest -n auto — parallel run green
• uv run --with pytest-randomly pytest -p randomly — green across multiple seeds (suite is order-independent)

[codecov]

Codecov Report

❌ Patch coverage is 69.23077% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.18%. Comparing base (fa740d3) to head (68d19d1).

Files with missing lines	Patch %	Lines
src/libvcs/pytest_plugin.py	66.66%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #539      +/-   ##
==========================================
- Coverage   61.19%   61.18%   -0.01%     
==========================================
  Files          40       40              
  Lines        6563     6557       -6     
  Branches     1103     1103              
==========================================
- Hits         4016     4012       -4     
+ Misses       1950     1948       -2     
  Partials      597      597              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[tony]

Code review

Found 1 issue:

1. The three new #### deliverable headings in the unreleased changelog block omit a PR ref (#539), unlike every other entry in the file (e.g. #### Command output is returned verbatim by default (#538)). (AGENTS.md says "PR refs (#NN) sit in each deliverable's #### heading.")

libvcs/CHANGES

Lines 23 to 42 in e918163

	### Fixes
	#### `VCSRegistry` no longer shares parser state across instances
	{class}`~libvcs.url.registry.VCSRegistry` kept its parser map in a class
	variable, so constructing any second registry mutated the shared
	{data}`~libvcs.url.registry.registry`. Each registry now owns its own parser
	map, leaving the global one untouched.
	#### `git_repo` and `hg_repo` fixtures return isolated clones
	The first consumer of the `git_repo` / `hg_repo` pytest fixtures received a live
	handle to the session-cached master checkout, so mutating that checkout (adding
	a remote, switching branches) polluted the cache for every later test. The
	master copy is now treated as a pristine, read-only cache and every consumer —
	including the first — gets its own copytree.
	### Development
	#### Opt-in parallel test runs

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}