[xiaoyuCR7]

Fixes #45166. _resolve_workflow_app_tenant_id is left to #43112, which already
handles it and has auto-merge enabled, so this PR does not touch it.

Summary

Message.workflow_run and WorkflowAppLog.workflow_run each build their own
sessionmaker from the global engine. I found no callers, so this PR removes
them instead of adding a session parameter.

• The console and service API app log endpoints serialize the DTOs built by
  WorkflowAppLogQueryRepository, not these ORM properties.
• The only log.workflow_run access I found, in the e2e tests, reads that DTO.
• No response schema reads workflow_run from a Message.
• The db import in models/model.py became unused, so it is removed, along
  with the models.model.db patch in test_human_input_form.py.

Provider credentials

Provider.credential and ProviderModel.credential were cached_property
accessors that opened their own session. ProviderManager already preloads these
rows in batches, so that lookup was only a hidden fallback.

• Provider.credential now reads a value attached by the caller through
  attach_credential and never queries. It raises RuntimeError if the credential
  was not attached, so a missing load cannot look like an unset credential.
• ProviderModel.credential, credential_name and encrypted_config had no
  readers and are removed. The loaded row is passed to
  _ProviderModelCacheEntry.from_record directly.
• _init_trial_provider_records could return an existing record from its
  IntegrityError branch without the batch load, so that branch attaches the
  credential too. This keeps the previous behavior for such a record.

Behavior change to review: reading a credential on a Provider loaded outside
ProviderManager without attaching now raises instead of querying. I found no
such reader in production code.

@GYJ99 I saw your comment on the same two accessors after preparing this
change. I'm opening it because the audit suggests they can be deleted. Happy to coordinate if you already have work in progress.

Testing

Run locally on Windows with Python 3.13, without the vdb-all dependency group:

• pytest on tests/unit_tests/models, core/entities, core/ops, core/app/llm,
  core/test_provider_manager.py, and the human input and workflow log tests:
  1061 passed, 1 error. The error is a Windows temp file PermissionError in
  test_trace_entity.py, and it also occurs on the unmodified base commit.
• Removing the new attach call from the IntegrityError branch makes its
  regression test fail with the expected RuntimeError.
• pyrefly check on the four changed source files: 0 diagnostics.
• lint-imports: 76 contracts kept, 0 broken.
• ruff format and ruff check pass on the changed files.

Not verified:

• The full unit test suite and make lint && make type-check were not run.
• Local runs used --no-cov and a stub for python-magic, because importing
  libmagic segfaults on this machine. CI is awaiting maintainer approval.

Screenshots

N/A, no UI change.

Checklist

• This change requires a documentation update, included: Dify Document
• I understand that this PR may be closed in case there was no previous discussion or issues. (This doesn't apply to typos!)
• I've verified the change and added or updated tests where meaningful regression risk justifies coverage.
• I've updated the documentation accordingly.
• I ran make lint && make type-check (backend) and vp staged (frontend) to appease the lint gods

From Claude Code

🤖 Generated with Claude Code

[GYJ99]

Thanks for the heads-up, @xiaoyuCR7. My earlier comment on #45166 was a proposed scope and test plan; I have not started a competing implementation. Your call-site audit covers the same Message.workflow_run and WorkflowAppLog.workflow_run accessors, so I won't open a duplicate PR. Removing truly unused properties looks preferable to adding a session parameter solely to preserve them. I appreciate you documenting the behavior change for provider credentials and the actual test limitations. I'll follow the review and help with focused verification if there's an uncovered case. Thanks for coordinating!

[Infernoooooo7]

Hey! I tested your changes locally and added two additional regression tests in test_provider_manager.py.

They cover:

• Handling deleted/missing provider credentials through _get_all_providers().
• Resolving active, dangling, and absent credentials in _load_provider_model_cache_entries().

The full provider manager test file passes locally (57 passed).

Would these additional tests be useful for your PR? I'd be happy to contribute them.