Vellum-ai

- Supports pulling Workflows without needing a `pyproject.toml` config defined.
- Uses a reasonable Workflow Sandbox label-inspired default module name on initial pull of a Workflow.

- Fix retry node serialization without explicit display class

- Code Execution Nodes in the SDK can now run inline instead of in Vellum when used without packages and defined as a Python Runtime
- Fixes a bug where External Inputs were not submitted correctly on two nodes of the same External Input interface.
- Fixes an issue around pulling the same workflow twice creating redundant entries in the Vellum Lockfile
- Fixes a bug where code generation would output the wrong runtime for Code Execution Nodes
- Fixes a bug where node output reference pointers to non-existent subworkflow deployment outputs were causing attribute errors
- Removes the `__all__` from code generated `__init__.py` files
- Allows for code generation of Subworkflow Deployment nodes even if the Deployment is not found

- Code generation now handles failures from unable to find a deployment
- Adds a new `--workflow-sandbox-id` param to `vellum push` to allow another means of identifying a Workflow to push

- Add support for plaintext search filtering on the Documents list route

- Support default values for environment variable references
- Fixes serialization of Vellum Secrets