Fix concurrent clique table publication

[derluke]

Closes #1774.

• Allocate and publish the initial clique table before launching its OpenMP producer task, avoiding an unsynchronized read/write of the owning shared_ptr.
• Let the task mutate only the pre-published table; the existing signal/dependency still joins mutation before cut generation reads it.
• Exercise 25 sequential solves with concurrent initial-clique generation enabled.

Pre-commit passes. The C++ regression requires the cuOpt build environment and MIP datasets, so it is left for CI.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.