Unify OCR test code and simplify blank PNG fixture generation

[etherlink-intern]

Motivation

• Remove duplicated OCR-related test scaffolding and make blank PNG fixture creation deterministic across platforms that provide image APIs.
• Simplify error messages and image encoding steps to reduce fragile test behavior when generating fixture data.

Description

• Removed a duplicate #if OCR test block and consolidated the Vision-based OCR tests into a single conditional section to avoid repetition.
• Reworked blankPNGFixtureData to generate a small blank PNG via CGContext and CGImageDestination when Vision, CoreGraphics, and ImageIO are available, and to fall back to reading the base64 fixture file otherwise.
• Simplified the in-memory PNG creation by using a 100x100 white canvas and by removing text rendering and compression options, and updated related ConversionError messages.
• Adjusted preprocessor checks to align with available image APIs and made the image destination use a concrete public.png UTI when generating the blank fixture.

Testing

• Ran the test suite with swift test which executed SwiftMarkItDownTests and related fixtures, and the tests completed successfully.
• OCR-related tests were exercised under the image-capable configuration and the blank-PNG fallback path was validated via the base64 fixture on platforms without those image APIs.

---

Codex Task