Phase 3 — opt_einsum-style path planner + plan cache reuse

[scttfrdmn]

Roadmap phase tracker

This issue tracks trntensor's work on Phase 3 of the trnsci roadmap.

See the suite-level roadmap
for the full phase matrix and cross-project dependencies. A reader-friendly
version lives at trnsci.dev/roadmap/.

What this phase means for trntensor:
Multi-tensor contractions have exponentially many possible
orderings; the FLOP cost varies by 10–100×. opt_einsum solves
this on CPU; Phase 3 does it on NKI.

Done means:

• multi_einsum with optimize="optimal" / "auto" / "greedy" path selection.
• Contraction plans cached and re-executed without re-planning.
• Benchmarks on canonical multi-tensor patterns (MP2, CCSD intermediates, THC-compressed evaluations).
• Integration with trnblas Phase 3 fused kernels for the hot-path contractions.

Coordination

• Label: phase-3-perf — matches the same label in every sub-project.
• Cross-project dependencies, if any, are called out in the "Done means"
  list above. Link any PRs / child issues here for tracking.

Close this issue when all "Done means" items are satisfied.

[scttfrdmn]

Plan cache implemented in 9cb532e (clear_plan_cache, plan_cache_info). Path planner (#18) was done in a2d594c. Benchmark item deferred until hardware access.