Fix: Sanitize whitespace in $ref paths for OpenAI strict mode

[PaulyBearCoding]

Fixes #1679

Problem Statement

When using zodResponseFormat with Zod schemas that contain property names with whitespace characters, the generated JSON Schema produces $ref values and definition keys containing literal spaces. This causes validation failures when submitting the schema to the OpenAI API.

Root Cause

In src/_vendor/zod-to-json-schema/parseDef.ts line 134, the extract-to-root case joins path segments with underscores but doesn't sanitize the segments themselves. Path segments like "Thing With Spaces" preserve their internal spaces in the final $ref value.

The join('_') method only adds separators between array elements - it does not modify the content within individual elements.

Solution

Map over each path segment and replace all whitespace characters with underscores before joining:

```typescript
const name = item.path
.slice(refs.basePath.length + 1)
.map((segment) => segment.replace(/\s+/g, ''))
.join('');
```

The regex /\s+/g matches all types of whitespace: spaces, tabs, newlines, and Unicode whitespace characters.

Testing

Unit tests:

• All 18 tests pass (16 existing + 2 new)
• Both Zod v3 and v4 compatibility confirmed
• New test verifies $ref values and definition keys contain no spaces

Comprehensive edge cases validated:

• Multiple consecutive spaces
• Tabs and newlines
• Unicode spaces (nbsp, em space, etc.)
• Deeply nested structures
• Valid names with underscores/hyphens preserved
• Emoji and CJK characters
• Performance: 0.03ms per schema generation (1000 iterations)

No breaking changes:

• Valid identifiers (alphanumeric, underscores, hyphens) remain unchanged
• Only whitespace characters are affected

Changes

Modified files:

• src/_vendor/zod-to-json-schema/parseDef.ts (4 lines)
• tests/helpers/zod.test.ts (added comprehensive test)