UPSTREAM PR #26917: python: fix duplicate proto field detection to use field number instead of JSON name

[loci-dev]

Note

Source pull request: protocolbuffers/protobuf#26917

Summary

Fix duplicate proto field detection in the Python JSON parser to use resolved field identity instead of raw JSON key strings.

Root cause

_ConvertMessage() tracked seen fields by their raw JSON key string. Because every proto field with an underscore has two valid JSON representations (snake_case proto name and camelCase json_name), a payload containing both aliases passed the duplicate check undetected.

Fix

# Before
if name in names:
    raise ParseError(...)
names.append(name)

# After
if field.number in names:
    raise ParseError(...)
names.append(field.number)

Field numbers are unique per message and independent of JSON representation, so both aliases for the same field are correctly detected as a duplicate.

Conformance

This brings the Python implementation into conformance with the proto3 JSON specification:

"Parsers are required to reject JSON that contains the same field (either by proto field name or by JSON name) more than once."

The C++ and Go implementations already handle this correctly.