whisper: add carry_initial_prompt to maintain context over sliding wi…

[Sahith59]

Description

This PR resolves the issue where context provided by the user's initial_prompt is lost/truncated as the cross-attention context window slides forward over long audio files.

Changes

• Introduces the carry_initial_prompt keyword boolean to .transcribe().
• Intercepts the prompt creation block inside transcribe.py. When enabled, it dynamically prepends the initial_prompt_tokens directly to the active prompt queue.
• Calculates the window boundary natively via model.dims.n_text_ctx to correctly slice the previous_tokens so that the model evaluates without hitting dimension mismatch errors during Apple MLX evaluation.
• Added native Apple MLX unit tests into test.py to test context-window bounds and propagation.
• Default behavior rigorously maintains PyTorch backward compatibility (carry_initial_prompt=False).

Testing

Ran local conversion via setUpClass and verified native propagation logic inside the test suite. Formatted with black and passes pre-commit natively.

Resolves: #1410