feat: warn on large text detector inputs

[hoobnn]

What

Warn when the actual text detector input produced by preprocessing exceeds 4,000,000 pixels.

• Uses the detector input dimensions after resize, without assuming whether the source was downscaled.
• Reports both detector and source image dimensions and gives pipeline/model-level sizing guidance.
• Emits at most once per processor instance, keeping deduplication state bounded.
• Keeps the warning policy in a small helper with focused boundary, deduplication, and message-context tests.

Verification

• python3 -m pytest -q -p no:cacheprovider tests/inference/models/text_detection/test_processors.py
• Result: 3 passed
• git diff --check

[paddle-bot]

Thanks for your contribution!