Add locked subgraph materialization primitives

[zsol]

uv check and uv format need to install the exact ty or ruff package already selected in uv.lock, including its source and transitive dependencies, without invoking the resolver again or modifying the project environment. The immediate follow-up is #19884.

This adds Lock::to_resolution, which materializes a Resolution from explicit locked Package identities. It preserves marker-selected universal forks, requested extras and dependency groups, install filtering, and shared dependencies across multiple roots, while excluding unrelated lock-manifest requirements.

CachedEnvironment::from_resolution ensures an environment for an existing Resolution and installs it without resolving again. from_spec now delegates to this path after resolution. Callers must supply the interpreter from which the resolution’s markers and tags were derived.

Concrete roots establish a new conflict context: conflict items inside the resulting subgraph are evaluated from the selected roots, extras, and groups. Materialization fails instead when a reachable dependency still depends on a conflict item outside that subgraph.

This change is a pure refactor, there should be no behavior change.

[astral-sh-bot]

uv test inventory changes

This PR changes the tests when compared with the latest main baseline.

• Added tests: 6
• Removed tests: 0
• Changed suites: 1

uv-resolver: +6 / -0

Added:

• uv-resolver::lock::installable::tests::installable_to_resolution_preserves_node_overrides
• uv-resolver::lock::installable::tests::materializes_conflicting_extras_within_the_synthetic_root
• uv-resolver::lock::installable::tests::materializes_multiple_concrete_roots_with_shared_dependencies
• uv-resolver::lock::installable::tests::materializes_the_selected_universal_lock_fork
• uv-resolver::lock::installable::tests::rejects_conflicts_outside_the_synthetic_root
• uv-resolver::universal_marker::tests::has_conflict_marker

Removed: none

[zanieb]

I know we have some external snapshot files, but preferably they're only for very large snapshots. Everything else should be inlined.

[zanieb]

I think Charlie is probably the best reviewer for this when it's ready. I can try, but it's a struggle.

[zsol]

Hey, it was a struggle for me too!

[zanieb]

Why not take a universal resolution (i.e. lock) and an interpreter and convert to the resolution here (you could even just call the method on Lock still) so a caller can't mess up this invariant?

[zanieb]

(for a second I thought the pull request was tiny then realized installable.rs was collapsed, I could learn why it's this way after reading that file)

[charliermarsh]

I thought we determined at some point in the past that it wasn't feasible to allow installation from an arbitrary root -- it had to be a workspace member. I can't find a comprehensive discussion though, just:

• #5008 (comment)
• #10082 (comment)

I think the concern was that around conflicting extras. We need to know which extras were activated that led to including a given package in order to evaluate its subgraph. But maybe that's not relevant here.