✨ Implement compute config reconciliation for VirtualMachine

[hpannem]

✨ Implement compute config reconciliation for VirtualMachine spec

What does this PR do, and why is it needed?

This PR introduces a unified compute configuration reconciliation path for
VirtualMachine objects. Previously, compute fields (CPU topology, resource
allocation, CPU/memory flags, latency sensitivity) were either not managed or
applied inconsistently across powered-on and powered-off reconfigure paths.

Key changes:

• OverwriteSpecComputeConfig — new core function using a computeFieldDef
  function-table pattern. Each field carries its own minHWVer, hotPluggable,
  differs, and apply callbacks. A 3-way comparison (spec vs. live vSphere
  state vs. class-derived ConfigSpec) determines whether to emit a change.
  hwVer is derived internally from liveCI.Version rather than passed as a
  parameter.

• Spec semantics — VM Operator fully owns every compute field it exposes.
  nil means "reset to platform default." The only exception is
  resources.size.cpu/memory where nil defers to the VirtualMachineClass.
  0 is a sentinel for coresPerSocket / vnumaNodeCount meaning "auto."
  -1 is a sentinel for limits.cpu / limits.memory meaning "unlimited."

• API validation (api/v1alpha6, webhooks) —

  • coresPerSocket and vnumaNodeCount minimum changed to 0 (allow "auto"
    sentinel).
  • XValidation rule tightened: vnumaNodeCount > 0 requires
    coresPerSocket > 0 (an explicit non-zero value).
  • -1 allowed for limits.cpu and limits.memory as the "unlimited"
    sentinel; 0 is still rejected for limits.

• SyncClassComputeToSpec / SyncClassSizeAndAllocationToSpec — new
  functions that copy the class's compute fields into vm.Spec when a
  class-based resize is triggered (ResizeNeeded returns true). This restores
  backward compatibility: during a resize the class is authoritative and
  overrides all spec fields (backfilled or user-specified) that it defines.
  The sync is performed in the reconciler — not the mutating webhook — because
  the VirtualMachineClass may not exist yet when the webhook fires.

  • Full class resize (VMResize feature): SyncClassComputeToSpec syncs all
    compute fields, and device changes are applied atomically in the same
    vSphere reconfigure call. The subsequent reconcile is a no-op.
  • CPU/memory-only resize (VMResizeCPUMemory feature):
    SyncClassSizeAndAllocationToSpec syncs only size and allocation fields,
    matching the narrow scope of that path.

Which issue(s) is/are addressed by this PR? (optional):

Fixes #

Are there any special notes for your reviewer:

• All new sync calls are gated by TelcoVMServiceAPI feature flag since
  OverwriteSpecComputeConfig only runs when that flag is enabled.

Please add a release note if necessary:

VirtualMachine compute fields (CPU topology, resource allocation, CPU/memory
flags, latency sensitivity) are now fully reconciled via `vm.Spec`. Setting a
field to `nil` resets it to the platform default. `0` is an "auto" sentinel for
`coresPerSocket`/`vnumaNodeCount`; `-1` is an "unlimited" sentinel for
`limits.cpu`/`limits.memory`. Class-based resize continues to be authoritative:
when a resize is triggered the class overrides all affected spec fields.

<!-- readthedocs-preview vm-operator start -->
----
📚 Documentation preview 📚: https://vm-operator--1644.org.readthedocs.build/en/1644/

<!-- readthedocs-preview vm-operator end -->