Fix memory leak with oneofs and PB_ENABLE_MALLOC

[Mifacopy]

Summary

This change backports/adapts the behavior from nanopb commit 4fe23595732b6f1254cfc11a9b8d6da900b55b0c. It fixes a memory leak in the decode path when using oneof fields together with PB_ENABLE_MALLOC.

Problem

When:

• PB_ENABLE_MALLOC is enabled,
• A message contains a oneof field,
• The oneof contains a static submessage with pointer fields,
• Decoded data contains two values for that submessage,

the previous logic in pb_release_union_field() did not free the old submessage correctly, and decode_static_field() then memsets the uninitialized memory regardless of whether it should be reset or not, leading to a leak.

Fix

Adjust the logic so that memset() is only performed when the which_field tag is different. This ensures previously allocated memory is released correctly and no leak occurs.

Impact

• Addresses a specific leak scenario during decoding.
• No other behavioral changes intended.

Reference

• Upstream commit: 4fe23595732b6f1254cfc11a9b8d6da900b55b0c