fix(ocr_client): default to 'Text Recognition:' prompt in ollama_generate mode

[mufradhossain]

Problem

When using api_mode: ollama_generate, the SDK sends requests to Ollama
with an empty prompt. Ollama requires a non-empty prompt to trigger the
model — without it, the model returns empty content and json_result is [[]].

Root Cause

The SDK builds the prompt from message content, but GLM-OCR's internal
pipeline sends image-only requests with no text. This works against the
native GLM-OCR API but Ollama silently ignores empty prompts.

Fix

Default to "Text Recognition:" when the prompt is empty, matching
Ollama's official documented usage for GLM-OCR.

Note: any non-empty prompt works (even "."), but "Text Recognition:"
is used as it matches the official Ollama docs for this model.

Fixes #207

[JaredforReal]

@mufradhossain unit tests failed, can u update the unit test for this scenario?

[mufradhossain]

The test update is already in the second commit 7ddd841. It updates both test_empty_messages and test_no_messages_key to expect "Text Recognition:". The CI is pending your workflow approval to run.

[JaredforReal]

LGTM, thanks! @mufradhossain