Deduplicate generated schema artifacts using pydantic:module annotation

## Context

The `schema_dedup` branch moved shared types (creators, ToolShedRepository, StepPosition, HasUUID, etc.) into `schema/common/common.yml` — but the codegen still expands them inline into every generated module. This means `gxformat2.py` and `native.py` each contain their own copy of every common type, and `_conversion.py` has duplicate conversion functions to marshal between structurally-identical-but-different classes.

## Proposal

Add a `pydantic:module` annotation to `schema-salad-plus-pydantic` so types can declare which Python module they belong to. When generating a module, types annotated with a different `pydantic:module` emit an `import` instead of a class definition.

### Schema side

```yaml
# common.yml
- name: StepPosition
type: record
pydantic:module: "gxformat2.schema.common"
fields: ...
```

### CLI side

New `--module-name` arg so the generator knows which module it's building:

```bash
# Generate common module
schema-salad-plus-pydantic generate schema/common/common.yml \
--module-name gxformat2.schema.common -o gxformat2/schema/common.py

# Format-specific modules import common types instead of inlining
schema-salad-plus-pydantic generate schema/v19_09/workflow.yml \
--module-name gxformat2.schema.gxformat2 -o gxformat2/schema/gxformat2.py
```

### Application side

- Import common types from `gxformat2.schema.common` instead of aliased duplicates from both modules
- Collapse duplicate conversion functions (`_convert_creators` / `_convert_creators_to_native`, `_convert_tool_shed_repo_to_format2` / `_convert_tool_shed_repo_to_native`)

## Open questions

- Abstract types in common module — emit as abstract BaseModel or skip?
- Strict variants — separate `common_strict.py` or share one common module?
- `load_document()` / `_load_single()` require document roots — common module should skip (library mode)?
- `pydantic:` namespace declaration in common.yml (currently only declared in importing workflow.yml files)
- Traditional codegen (`v19_09.py`, `native_v0_1.py`) — leave as-is or drop? Nothing imports them.