Reuse the existing arm_ml_service generated client for the v2025_01_01_preview
API surface instead of importing from the per-version restclient package, and
delete azure/ai/ml/_restclient/v2025_01_01_preview/.

- _ml_client: wire the v2025_01_01_preview service client via a partial of
  MachineLearningServicesMgmtClient with api_version='2025-01-01-preview'.
- Switch production imports (command builder/entity, job_resource_configuration,
  to_rest_functions, capability_host entity + operations) from
  azure.ai.ml._restclient.v2025_01_01_preview to
  azure.ai.ml._restclient.arm_ml_service.
- command_job.py: the shared CommandJob hybrid model does not declare
  parent_job_name, but it is a real field on the 2025-01-01-preview wire
  contract (wire key 'parentJobName'). Preserve it via dict-style assignment on
  the rest model rather than passing it as a constructor kwarg, and read it back
  via dict access in _load_from_rest.
- job_resource_configuration.py: same situation for max_instance_count
  (wire key 'maxInstanceCount', still declared on v2023_04_01_preview). Set it on
  the wire via dict assignment on the 202501 branch and read it back via
  obj.get('maxInstanceCount') for the hybrid model. Verified the value survives
  serialization and round-trip.
- Update conftest mock fixture and capability host entity test to import from
  arm_ml_service.
- test_command_job_schema: environment_variables is Dict[str, str] on the REST
  contract; the shared client coerces values to strings, so assert against the
  stringified expected values.
- Delete the now-unused azure/ai/ml/_restclient/v2025_01_01_preview/ package.

…n path

Verified against the frozen 2025-01-01-preview swagger (mfe.json on the
azure-ai-ml/before_tsp branch): maxInstanceCount does not exist anywhere in that
contract (0 occurrences; absent from JobResourceConfiguration and its base
ResourceConfiguration). The old v2025 autorest model silently ignored the
max_instance_count kwarg and never serialized it, so the wire-identical behavior
is to OMIT it on the arm_ml_service (202501) branch. The previous dict-style
assignment added an off-contract field and is reverted. The v2023_04 (else)
branch still sets max_instance_count because that version's swagger declares it.

parentJobName is unaffected and remains preserved via dict assignment in
command_job.py, since it IS a real 2025-01-01-preview field (declared on JobBase).

…ient

Consolidate the separate v2024_10_01_preview_tsp TypeSpec client onto the shared
arm_ml_service client (api_version='2024-10-01-preview') and delete the
v2024_10_01_preview_tsp package.

- _ml_client: wire the 2024-10 client as a partial of MachineLearningServicesMgmtClient
  with api_version='2024-10-01-preview' (kept the ServiceClient102024PreviewTsp alias name).
- Flip all 21 production imports + 7 test files from v2024_10_01_preview_tsp to arm_ml_service.
- THROWAWAY SHIM for two workspace properties that the shared client no longer declares:
  allowRoleAssignmentOnRG and networkAcls were @removed at api-version 2025-12-01 in the
  TypeSpec, and the shared client is generated at 2025-12-01, so its Workspace /
  WorkspaceUpdateParameters hybrid models reject them as kwargs. Both are still valid on the
  2024-10-01-preview wire (verified in workspaceRP.json). Added
  set_removed_workspace_property / get_removed_workspace_property helpers in
  _utils/_workspace_utils.py that round-trip these through the hybrid model's flattened
  'properties' envelope. networkAcls is nested and serialized via NetworkAcls.as_dict().
  Remove this shim once the shared client is regenerated with api-version 'all'.
- Update test_workspace_operations_base to assert the wire value
  (params.as_dict()['properties']['allowRoleAssignmentOnRG']) instead of the dropped attribute.
- Delete the now-unused azure/ai/ml/_restclient/v2024_10_01_preview_tsp/ package.

Move the workspace-connection auth models off the per-version v2022_01_01_preview
package onto the shared arm_ml_service client and delete the package.

- entities/_credentials.py: re-source the workspace-connection auth models from
  arm_ml_service with their current (renamed) names:
  PersonalAccessToken -> WorkspaceConnectionPersonalAccessToken,
  ServicePrincipal -> WorkspaceConnectionServicePrincipal,
  SharedAccessSignature -> WorkspaceConnectionSharedAccessSignature,
  UsernamePassword -> WorkspaceConnectionUsernamePassword,
  ManagedIdentity -> WorkspaceConnectionManagedIdentity.
- Cross-version coupling fix: the compute identity (IdentityConfiguration
  _to_compute_rest_object / _from_compute_rest_object, used by aml/compute_instance/
  kubernetes/synapsespark compute) is serialized through the msrest v2022_10
  ComputeResource. Source those two models from v2022_10_01_preview msrest instead
  of arm_ml_service so msrest serialization keeps working:
  Identity -> v2022_10_01_preview.ManagedServiceIdentity (same fields), and
  UserAssignedIdentity -> v2022_10_01_preview.UserAssignedIdentity. Without this the
  hybrid models raise SerializationError ('object has no attribute _attribute_map').
- Update conftest mock fixture and the connection unit test to import from
  arm_ml_service.
- Delete the now-unused azure/ai/ml/_restclient/v2022_01_01_preview/ package.