[LauraGPT]

Summary

Release the already-merged #3838 Contextual/Seaco Paraformer hotword batching and empty hotword-file fix as standalone funasr-onnx 0.4.7. Upgrading the root funasr package alone does not update this separate distribution.

This PR changes only the standalone version, README, package workflow version/artifact name, and the two existing version-contract tests. No inference code, dependency range, model weight, root package version, or repository protection changes.

Validation

• Updated release assertions first: 3 expected failures and 5 passes against the old metadata, then all 8 contracts passed after the release changes.
• Built wheel and sdist, passed Twine checks, and installed the wheel into an isolated target; the installed version/origin test passed.
• All 13 packaged runtime modules match the accepted fix(runtime): decode every batch in ContextualParaformer and SeacoParaformer #3838 merge-source bytes. Reused the existing 52 selected CPU tests and 48 real NVIDIA L20Z FP32 analytic TorchScript/ONNX batching contracts (1,850 Torch CUDA events, 590 ONNX CUDA nodes, no CPU fallback nodes). These are previously completed runtime-contract tests, not new inference experiments or pretrained ASR quality/full-suite results.
• Exact new-head CI and actual hosted package artifacts must pass before normal merge. New main, tag, GitHub asset bytes and PyPI payload bytes are separately gated after merge.

Existing limitations remain: no new pretrained ASR accuracy, Seaco timestamp-output, Windows, NPU, or full-suite claim. Nano work remains deferred. Publication has not occurred merely because this PR exists.