[Docs] Update video caption model matrix and Qwen migration

[lbliii]

Context

Child of #2118. #1827 added Qwen 3 support and replaced the generic qwen variant with qwen2.5 / qwen3. #1883 added Nemotron 3 Nano Omni captioning. The published caption page still uses qwen, omits the Omni variant, and conflates the caption-enhancement algorithm with the separate enhanced-model selector.

This is a copy-paste failure against the current CLI and Python APIs.

Required coverage

• Document qwen2.5 and qwen3 model variants, model IDs, defaults, and preprocessing differences.
• Add nemotron-3-nano-omni to the model matrix and examples.
• Replace every stale model_variant="qwen" and --captioning-algorithm qwen example.
• Document current --captioning-algorithm, --enhance-captions-algorithm, and --enhanced-caption-models choices without conflating them.
• Add migration guidance from qwen to qwen2.5.
• Coordinate with [video] Remove model_does_preprocess #2109 so no example restores model_does_preprocess or --captioning-model-does-preprocess.
• Verify caption preparation, generation, enhancement, prompt-format, and output-key behavior for each model family.

Target pages

• fern/versions/main/pages/curate-video/process-data/captions-preview.mdx
• Related video overview/tutorial pages and release migration guidance

Acceptance criteria

• Every documented CLI choice matches the current parser
• Every Python example uses a valid current model variant
• Qwen 2.5, Qwen 3, and Nemotron 3 Nano Omni have accurate setup and selection guidance
• The qwen → qwen2.5 breaking change has a concrete migration example
• Caption enhancement flags and model selectors are clearly distinguished
• Removed preprocessing controls do not appear in runnable examples
• Fern validation and link checks pass

Related PRs: #1827, #1883, #2109