Python: add an option to keep valid original property names

## Summary

The Python renderer's default `nice-property-names` behavior splits valid identifiers at digit/letter boundaries. For example, the JSON property `source_m3u8` becomes the Python field `source_m3_u8`.

This is surprising for protocol/API fields and causes a practical problem with the existing `--pydantic-base-model` option: FastAPI/Pydantic validates request bodies against the generated Python field name, but the JSON payload still contains `source_m3u8`.

## Reproduction

Input schema:

```json
{
"type": "object",
"properties": {
"source_m3u8": { "type": "string" }
},
"required": ["source_m3u8"]
}
```

Generate Python:

```bash
quicktype \
--src schema.json \
--src-lang schema \
--lang py \
--python-version 3.7 \
--pydantic-base-model \
--top-level Request
```

The generated model contains:

```python
class Request(BaseModel):
source_m3_u8: str
```

Parsing the real API payload fails:

```python
Request.model_validate({"source_m3u8": "https://example.test/live.m3u8"})
# ValidationError: source_m3_u8 - Field required
```

The generated `from_dict` helper does reference `obj.get("source_m3u8")`, but FastAPI/Pydantic does not call that helper when validating a request model.

Using `--no-nice-property-names` is not an adequate workaround because it produces `sourcem3u8`, which still does not preserve the original property name.

## Proposed behavior

Add an opt-in Python renderer option, for example `--keep-property-names`:

- default remains `false`, preserving current output;
- when enabled, keep the original name if it is a valid Python identifier;
- continue using the existing naming/legalization behavior for invalid identifiers and Python keywords.

This keeps the change backwards compatible while allowing generated Pydantic models to match API JSON keys directly.

Start at the Python renderer and CLI option handling, then reproduce the schema with the quicktype command shown, including --pydantic-base-model. Done means an opt-in --keep-property-names preserves valid original identifiers such as source_m3u8, while invalid identifiers and Python keywords retain existing legalization and default output remains unchanged.

Written by the indexing model from the issue text.