Reservations capacity assesment

[jamOne-]

Description

The change enhances how XPK handles reservations during cluster creation.

Important: if the new validation logic is blocking you in an unexpected way, please let us know!
To disable the new behavior use the RESERVATIONS_VALIDATION_ENABLED=False environment variable when running xpk:

RESERVATIONS_VALIDATION_ENABLED=False xpk ...

---

Key Improvements for Users:

• Smarter Reservation Allocation: Instead of blindly cycling through provided
  reservations, XPK now dynamically assesses the actual available capacity (slices/blocks)
  within each reservation. It allocates these available slots to the requested node pools,
  ensuring better utilization.
• Pre-flight Validation: XPK now validates that your reservations match the requested
  system configuration (e.g., correct machine type, accelerator type) and have sufficient
  healthy capacity before starting the cluster creation process. This prevents failures
  later in the process.
• Enhanced Super-Slicing Support: For super-slicing workloads, the tool can now
  intelligently identify and target healthy sub-blocks within a reservation block.

---

Next step: we'll make --num-slices / --num-cubes optional when using the reservation capacity type.

Issue

• http://b/477114712
• http://b/485645043
• http://b/477114716

Testing

Manual testing scenarios:

1. New cluster, super-slicing, 3 slices, reservations passed: subblock, block.
2. Existing cluster with 3 slices, super-slicing, +1 slice, reservations passed: partially taken subblock, subblock (xpk should fail).
3. Existing cluster with 3 slices, super-slicing, +1 slice, reservations passed: subblock.
4. Existing cluster with 4 slices, super-slicing, +1 slice, reservations passed: reservation.
5. New cluster, GPU, --num-nodes=9999, reservations passed: GPU reservation (xpk should fail).
6. New cluster GPU slice, reservations passed: TPU reservation (xpk should fail).
7. New cluster GPU, 2 nodes, reservations passed: GPU reservation.
8. New cluster TPU, 2 slices, reservations passed: TPU reservation.
9. New cluster TPU, 2 slices, reservations passed: different TPU machine reservation.
10. New cluster CPU reservations.