[LauraGPT]

Scope

Release the already merged standalone ONNX wrapper fixes as funasr-onnx==0.4.11:

• fix(runtime): trim batch padding from CIF peaks before funasr_onnx and funasr_torch Paraformer timestamps #3871: timestamps do not inherit batch padding from longer inputs.
• fix(runtime): cut long hotwords to 10 characters and skip empty ones #3882: Contextual/Seaco hotwords stay within the exported ten-character embedding width; whitespace-only entries are skipped.
• fix(runtime): merge spelled-out acronyms into one word in funasr_onnx and funasr_torch sentence_postprocess #3869: terminal abbreviations retain text and combined timestamps.

Only release metadata, README and installed-package contracts change. Runtime modules match current main; no new weights or dependency changes.

Verification

• Actual official 0.4.10 wheel: three new regression methods fail as expected.
• New source release/postprocessing checks: 60 passed, no skips.
• Separate new wheel and sdist installations: 45 passed each, no skips; strict Twine check passed.
• All 13 packaged runtime modules match the existing accepted fix(runtime): cut long hotwords to 10 characters and skip empty ones #3882 CUDA candidate by bytes. Its bounded FP32 synthetic validation covered 24 backend/wrapper/batch cases and 72 outputs from three audio fixtures, including real L20Z execution and archived CUDA profiles. Completed GPU tests are reused by source hash, not rerun.
• The new hosted Python 3.11/3.12 matrix must additionally prove clean torch-free package installation before merge.

Synthetic contracts are not pretrained ASR quality, BF16, full-suite or Nano validation. GitHub Release/tag/assets and PyPI publication remain separate gates after merge; root 1.4.26 stays unchanged.