fix(protocol): Flatten Linux distribution fields into os.context

[JoshuaMoelans]

(continuation of #3443 )

To make these fields searchable, we had to change from a nested to a flattened approach. This is provided in sentry-native (relevant PR).

The update is also tracked on the docs side: getsentry/sentry-docs#11936

[JoshuaMoelans]

I'm not sure if this is a skip-changelog kind of situation; the previous (non-flattened) implementation is listed as a feature under 24.4.1 , so maybe this update should be labelled as a fix/internal?

[jjbayer]

Why does it need to be flattened to become searchable? Could we fix this on the storage end instead of changing the schema?

[JoshuaMoelans]

@jjbayer snuba cannot handle the nesting we had before (and the team apparently was reluctant to add this feature, hence this PR was closed without merging). They suggested to flatten our 3-layered context values back to 2 layers.

[jjbayer]

snuba cannot handle the nesting we had before (and the team apparently was reluctant to add this feature, hence this PR was closed without merging). They suggested to flatten our 3-layered context values back to 2 layers.

@JoshuaMoelans have SDKs already been sending the nested format? In that case, we need to still accept the nested format and normalize it into the flat format.

[JoshuaMoelans]

@jjbayer no, since we never merged & released the sentry-native PR we have not been sending these nested format events.

[jjbayer]

@jjbayer no, since we never merged & released the sentry-native PR we have not been sending these nested format events.

In that case I'm running out of objections 😄. Please add a **Breaking Changes**: entry to both CHANGELOG.md and py/CHANGELOG.md because this is technically a breaking change to the protocol, even if none of our own SDKs was using it yet