Preserve key type for Optional defaults

[gaoflow]

Summary

Optional(key, default=...) stringifies the key, so when the optional key is absent the default fills the output dict under str(key), while when the key is present the original-typed key is used. The output key type silently depends on whether the data contained the key.

from schema import Schema, Optional

s = Schema({Optional(5, default="x"): str})
s.validate({})        # -> {'5': 'x'}   (string key!)
s.validate({5: "p"})  # -> {5: 'p'}     (int key)

The same affects None (→ 'None'), float, and bool keys. String keys are unaffected (str("a") == "a"), which is why the existing tests — all using string keys — never caught it.

Cause

Optional.__init__ set self.key = str(self._schema). The default-fill path then writes new[default.key] = ..., so the stringified key reached the output. A present key, by contrast, flows through normally and keeps its type.

Fix

Store the original key object (unwrapping a Literal, whose str() form the old code relied on for description/title keys), mirroring Hook.key = self._schema:

self.key = (
    self._schema.schema
    if isinstance(self._schema, Literal)
    else self._schema
)

Now present and absent paths produce the same key type. Literal and plain string keys are unchanged.

Verification

• Added test_dict_optional_default_preserves_key_type, asserting that int/None/float keys keep their type when filled from the default and that present/absent agree. It fails before the fix and passes after.
• Full suite: 121 passed (was 120, no pre-existing failures). ruff check adds no new errors (the pre-existing B023 reports in json_schema are untouched); ruff format --check reports already-formatted.

This pull request was prepared with the assistance of AI, under my direction and review.