Add TensorRT-LLM Transfer, Action, and Distilled guides

[ishovkun]

Summary

• add a TensorRT-LLM Transfer walkthrough for edge, blur, depth, segmentation, and WSM controls
• add TensorRT-LLM Action walkthroughs for forward dynamics, inverse dynamics, and DROID policy inference, including the tensor/action response contract
• extend the TensorRT-LLM audiovisual notebook with the fixed-schedule four-step T2I and I2V checkpoints
• update the root index, shared setup, and workflow READMEs with concise launch and capability guidance

Validation

• parsed all five affected TensorRT-LLM notebooks as JSON
• verified unique cell IDs, empty outputs, null execution counts, and Python compilation for every code cell
• verified referenced checked-in assets, local Markdown links, and fenced blocks
• exercised multipart construction and encoded-video/tensor decoding with mocked Transfer and Action responses
• verified distilled requests omit checkpoint-owned step, guidance, and I2V system-prompt overrides
• ran git diff --check

GPU inference was not run because the final TensorRT-LLM Transfer and Action implementations are not merged yet. The request/response contracts were validated against their current reviewed implementations and mocked responses; representative GPU runs remain required after those implementations land.

Dependencies and follow-up

This PR is stacked on #307 and should be rebased onto main after that PR lands, preserving both guide sets.

Keep this PR in draft until NVIDIA/TensorRT-LLM#16394 and NVIDIA/TensorRT-LLM#17325 merge. Then recheck their final APIs, rebase, rerun the local validation above, and perform representative GPU inference before marking this ready for review.