Add scheduling.provisioningModel per-process hint for Google Batch

[reece-maticebio]

Summary

closes #3530

Implements the scheduling.provisioningModel hint from the accepted hints-process-directive ADR
(20260323-hints-process-directive.md) for the Google Batch executor, resolving #3530. It lets a
pipeline choose spot / on-demand / preemptible provisioning per process (rather than only globally
via google.batch.spot), through the hints directive.

What it adds

• A scheduling.provisioningModel hint (a String) on the Google Batch executor, mapping to the
  Google Batch ProvisioningModel enum:
  • spot → SPOT
  • standard (alias: ondemand) → STANDARD
  • preemptible → PREEMPTIBLE
  • values are case-insensitive and trimmed.
• Strict validation. Any other value throws IllegalArgumentException — no silent fallback to a
  default, so a typo fails fast instead of quietly running on the wrong provisioning model.
• Unknown-key validation. An unrecognized google-batch/-prefixed hint key throws, per the ADR
  (unrecognized prefixed hints are errors), matching the Seqera executor. Bare and foreign-executor
  keys are left untouched.
• Resolution. The hint resolves both the bare key (scheduling.provisioningModel) and the
  executor-prefixed key (google-batch/scheduling.provisioningModel); the prefixed form is prioritized. This
  mirrors the AWS Batch executor's existing hint resolution (get(prefix+key) ?: get(key)).
• Fallback. When the hint is unset, provisioning is unchanged from today's global behaviour
  (google.batch.spot → spot, google.batch.preemptible → preemptible, else on-demand). STANDARD
  is applied by omitting the field so Batch's on-demand default takes effect. Pricing selection
  (PriceModel.spot) is derived from the resolved model so cost estimation stays consistent.

Tests

Unit tests added in nf-google:

1. should resolve provisioning model from scheduling.provisioningModel hint covers
   • Bare SPOT/STANDARD/PREEMPTIBLE`
   • google-batch/-prefixed form, prefixed-wins precedence,
   • case-insensitivity and trimming, the ondemand alias, and the unset global-fallback rows
2. should fail on invalid scheduling.provisioningModel hint value
   • a sweep proving strict value validation: non-enum strings, empty string, and non-String values all throw.
3. should throw on unknown google-batch prefixed hint
   • an unrecognized google-batch/-prefixed key (a realistic case-typo and a wholly unknown key) throws, per the ADR's prefixed-error rule.
4. should report all unknown google-batch hints together
   • two unknown prefixed keys produce a single exception naming both, proving errors are collected rather than thrown on the first.
5. should ignore unknown bare and foreign-executor hints
   • a bare key and a foreign-executor (awsbatch/…) key are left untouched (no throw), since an executor only validates its own prefixed keys.
6. should use instance template and ignore the provisioningModel hint
   • a template:// machine type with the hint set still submits successfully with the template honored (the hint is a no-op under an instance template, which replaces the instance spec wholesale).

Validation on real Google Batch

Exercised against a live Google Batch project:

• spot → VM provisioned SPOT
• standard and ondemand → STANDARD (on-demand)
• preemptible → PREEMPTIBLE
• no hint → falls back to the global google.batch.* setting
• invalid value → run fails fast with a clear error (no silent default)

Design Questions

A few points came up while I was working on this:

• Key name. The ADR catalogs this as scheduling.provisioningModel (what this PR implements), but
  the Seqera executor shipped machineRequirement.provisioning (via its
  SchemaMapperUtil.toProvisioningModel). I followed the ADR.
• Value vocabulary. Google's enum and the ADR use standard; AWS and Seqera use ondemand
  (Seqera also has spotFirst). This PR accepts both standard and ondemand.
• Unknown-hint handling differs from AWS. Unknown google-batch/* keys throw here (ADR line 114;
  matches Seqera). The AWS Batch executor currently only warns for unknown awsbatch/* keys, so the
  two Batch executors differ on this.
• Shared resolver (candidate follow-up). hints now has three specific consumers: AWS Batch
  (awsbatch/consumableResources), this Google impl (same get(prefix+key) ?: get(key) +
  warnUnknownHints pattern), and Seqera's HintHelper, which already generalizes prefix/namespace
  resolution behind extractSeqeraHints(). AWS and Google duplicate what HintHelper does. I'd love to take
  this on as a separate PR is deemed worthwhile. I kept it out here to stay scoped to Google Batch spot enabled configuration for processes #3530.

[netlify]

✅ Deploy Preview for nextflow-docs ready!

Name	Link
🔨 Latest commit	26d1aad
🔍 Latest deploy log	https://app.netlify.com/projects/nextflow-docs/deploys/6a580a3aac19d7000840d87a
😎 Deploy Preview	https://deploy-preview-7343--nextflow-docs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.