For continuous profiling (ProfileChunk), Relay infers the data category from the contents of the profile chunk.
SDKs are now sending the platform (platform) as an item header, which means Relay can now correctly determine the correct data category much earlier.
Relay needs to now consider the platform header where available to infer the the correct data category, but still fall back to the item contents (current behaviour) if the header is missing (old SDKs).
Example PR where this was added to the Python SDK: getsentry/sentry-python#4178
In a few months time, Relay can reject profile chunks which do not carry a platform in the item header.