Fix DecodingPress argument names in README

[fallintoplace]

Summary

• update the DecodingPress README examples to use the current constructor argument names
• fix the listed default values for compression_interval, target_size, and hidden_states_buffer_size
• clarify that the examples compress to a target cache size, not a token buffer size

Root cause

The README examples still used older argument names like compression_steps and token_buffer_size, while the implementation now expects compression_interval, target_size, and hidden_states_buffer_size. Copy-pasting the old examples raised TypeError.

Validation

• rg -n "compression_steps|token_buffer_size|compression_interval|target_size|hidden_states_buffer_size" README.md
• `uv run --python 3.11 python - <<'PY'
  from kvpress import DecodingPress, KnormPress, CriticalKVPress, PrefillDecodingPress

DecodingPress(base_press=KnormPress(), compression_interval=10, target_size=512)
PrefillDecodingPress(
prefilling_press=CriticalKVPress(KnormPress()),
decoding_press=DecodingPress(base_press=KnormPress(compression_ratio=0.2), compression_interval=5, target_size=256),
)
print("doc-snippets-ok")
PY`

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.