Keep workflow compression payloads self-describing

[lucasmrdt]

Context

This PR is stacked on #526. It fixes the workflow worker failure where existing callers still unpack encode_payload_content(...) as two values:

encoded_data, encoding_options = await encoder.encode_payload_content(data, context)

#526 introduced a third returned value for compression metadata, which breaks that caller shape with ValueError: too many values to unpack.

Changes

• Restores encode_payload_content(...) to its two-value contract: (encoded_data, encoding_options).
• Wraps compressed bytes in a CompressedPayloadData envelope containing:
  • the compression config
  • the base64-encoded compressed payload
• Decodes compressed payloads from that envelope, so compression metadata travels with the bytes instead of through a separate return value.
• Keeps the tampered-compression-level regression test: zstd decompression must not depend on the configured compression level.
• Raises WorkflowPayloadCompressionException for invalid compressed payload envelopes/base64/config instead of leaking generic validation errors.

Tests

• uv run ruff check src/mistralai/extra/workflows/encoding/payload_encoder.py src/mistralai/extra/workflows/encoding/payload_compressor.py src/mistralai/extra/tests/test_workflow_encoding.py
• uv run pytest src/mistralai/extra/tests/test_workflow_encoding.py -q