fix(normalization): Skip validation on renormalization

[iker-barriocanal]

Previous normalization PRs run validation from librelay calls. However, some sentry tests run full normalization in librelay on partial (i.e. invalid) payloads, and running validation makes plenty of tests fail (see getsentry/sentry#66134).

Similarly, many sentry tests expect span normalization to never happen. After spending significant time "fixing"* some tests, it'd require me even more time to "fix" the remaining ones as it requires a deeper understanding of other features/codebases. Thus, I'm reintroducing normalize_spans to disable span normalization (mostly exclusive time computation) from librelay calls and unblock the librelay release.

*Note: "fixing" a test means using data relay outputs from normalization. This doesn't align with other assumptions made in the tests, nor modifications to the event payload in the rest of the event processing pipeline.

Future work

• Consider span normalization a regular step in event normalization behind the is_renormalize flag, removing the normalize_spans flag.

[jjbayer]

Approving to unblock. It would be nice to get rid of the config flag eventually.

[Dav1dde]

I don't fully understand the change and it seems to me we're working around bad tests in Sentry, but we are definitely blocked on this.

Have you considered exposing the flag from librelay via Keyword arg, so we have at least a chance of fixing the Sentry tests on a step by step basis?

[iker-barriocanal]

@Dav1dde great idea! Exposed the flag in 380895a. It can be enabled by initing the StoreNormalizer with normalize_spans=True, is_renormalize=False.