fix(python): fill omitted template variables from prompt input.default

[huangjeff5]

input.default in a .prompt file fills any template key the caller left out. Call defaults overlay the file. Runtime input wins that key. render and compile take a DataArgument, not a plain dict. Model and config layers compose: instance default, then the file, then compile metadata, then the call.

from dotpromptz import Dotprompt
from dotpromptz.typing import DataArgument, PromptInputConfig, PromptMetadata

source = """---
model: gemini-2.5-flash
input:
  default:
    name: World
---
Hello, {{name}}!"""

await Dotprompt().render(source, DataArgument(input={}))
# model gemini-2.5-flash, text 'Hello, World!'

await Dotprompt().render(source, DataArgument(input={'name': 'Ada'}))
# 'Hello, Ada!'

await Dotprompt().render(
    """---
input:
  default:
    first: Ada
    last: Lovelace
---
{{first}} {{last}}""",
    DataArgument(input={'first': 'Grace'}),
)
# 'Grace Lovelace'

await Dotprompt().render(source, {'input': {'name': 'Ada'}})
# TypeError: data must be a DataArgument

Decisions

• DataArgument() and DataArgument(input={}) are both “no keys supplied.” File defaults fill.
• A key the caller did set, including None / False / 0 / '', replaces the default. Nested objects replace; they do not deep-merge.
• Call PromptMetadata(input=PromptInputConfig(default=…)) overlays the file for those keys. Compile metadata overlays the file; the call overlays compile; runtime input is last.
• The returned prompt config still carries the merged input.default map. That is the same overlay the template used.
• render and compile require DataArgument. A mapping is TypeError.
• Model: instance default_model, then file model:, then compile metadata, then the call. An empty string means absent, not a model named "".
• Config: the selected model’s model_configs entry, then file config:, then compile, then the call. Later keys overlay.
• YAML model: is kept on the parsed prompt.