Allow reuse of resolved Python test matrix snapshots

[vyasr]

Description

Add a reusable resolver that returns wheel-test and Conda Python-test matrices from one preparation of the shared matrix definitions. Callers can plan complementary coverage using a single snapshot, without copying these definitions into downstream repositories.

The wheel and Conda Python test workflows accept an optional resolved-matrix JSON input. Matrix validation, existing filters, and pure-package selectors remain centralized in the shared implementation. When no snapshot is supplied, callers retain the existing resolution behavior. Supplied snapshots do not reread mutable matrix definitions.

Regression tests cover default callers, filtering, pure-package selectors, duplicate matrix types, malformed inputs, and reuse despite changed source definitions. The resolver and consumer should use the same shared-workflows revision.

Pin the existing proxy-cache actions to their current revision to satisfy the repository's pinned security checks on the affected workflows.

[vyasr]

Closing in favor of implementing the complementary CUDA coverage policy entirely in cuDF using the existing compute-matrix and matrix_filter interfaces. We accept independently resolved matrix snapshots within a workflow run, so no shared-workflows API changes are needed.