Remove the transitional Job.compute_profile string field (step 1)

[Elena Peña Tapia (ElePT)]

Important

BLOCKED — needs a change from the billing team before this can merge.

The Kafka usage-event path builds the classical metric type from
job.compute_profile (the string this PR removes):
_build_classical_metric_type in
gateway/core/ibm_cloud/event_streams/kafka_event_streams_client.py.
That file is owned by another team, so this PR deliberately does not touch it.

Before this removal can merge, the billing team needs to repoint that method:

# gateway/core/ibm_cloud/event_streams/kafka_event_streams_client.py
-        if job.compute_profile:
-            parts.append(job.compute_profile)
+        compute_profile = job.compute_profile_id          # FK-backed, added in this PR
+        if compute_profile:
+            parts.append(compute_profile)

Job.compute_profile_id (added here) returns compute_profile_fk.compute_profile_id
(or None), which equals the old string wherever it was set — so the emitted
metric type is unchanged. Until that lands, merging this PR would raise
AttributeError when a usage event is built. The corresponding test
(tests/core/services/ibm_cloud/event_streams/test_event_streams_client.py) is
also left for them to update alongside their change.

Summary

Removes the transitional Job.compute_profile string field. It was introduced
alongside the t-shirt sizes work as a stopgap and superseded by Job.compute_profile_fk
(FK → ComputeProfile), which is the source of truth for the profile a job ran on. The two
were written together on every run and always agreed at creation; this PR retires the string.

This is Step 1 of a two-step removal that keeps every deployed release safe to roll back —
the same pattern used to drop Program.default_compute_profile (#2420 → #2421):

• This PR (Step 1): repoint every reader to the FK, then remove the field from Django's
  model state only — the api_job.compute_profile column is left in the database, so
  the previous release (whose Job model still declares the field and lists it in every
  SELECT) stays deployable.
• Step 2 (separate later PR): the real DROP COLUMN, once no release that still declares
  the field can be deployed.

What changed

Readers repointed to the FK. A new null-safe Job.compute_profile_id property returns
compute_profile_fk.compute_profile_id (or None for Ray / historical rows), preserving the
existing … or settings.DEFAULT_COMPUTE_PROFILE fallback everywhere. Repointed:

• Fleet-name construction and resource-limit parsing — core/services/runners/fleets_runner.py
• Admin job timeline — api/domain/job_timeline.py (+ select_related("compute_profile_fk")
  on the two timeline querysets in api/admin.py to avoid an N+1)

The billing/Kafka reader is intentionally not repointed here — see the blocked note above.

Writers. The string is no longer written on the run path (api/use_cases/programs/run.py)
or the filler-jobs path (scheduler/tasks/balance_filler_jobs.py); both already set
compute_profile_fk.

API shape. compute_profile (string) is dropped from the job serializers; compute_profile_fk
(nested ComputeProfileSerializer, already present on the list/retrieve responses) carries the
profile. Client Job loses its compute_profile attribute.

Admin. The compute_profile field is removed from the JobAdmin "Fleets" fieldset
(compute_profile_fk stays).

Migration. 0064_remove_job_compute_profile_state.py — a state-only RemoveField wrapped in
SeparateDatabaseAndState (empty database_operations), mirroring 0057. Its comment also
records the billing-team dependency above.

Out of scope

• The real DROP COLUMN (Step 2, follow-up PR).
• The billing/Kafka reader and its test (another team — the blocked note above is their action).
• The deprecated compute_profile run input kwarg on function.run(...) / the run
  InputSerializer — a separate surface, retired separately. function_size is the replacement.

Notes

• Invariant relied on: RunnerConfig.compute_profile is set to compute_profile_fk.compute_profile_id
  in every branch of _get_runner_config, and Ray leaves both null — so string and FK agree
  wherever the FK is non-null.
• Rollback-safe: the DB column is retained for one release.
• Related: internal t-shirt sizes epic; the Program.default_compute_profile precedent (Keep the default_compute_profile column so a previous release can still be deployed #2420/Drop the api_program.default_compute_profile column #2421).

Opened as draft for review.