feat(spans): Normalize and copy measurements to segments

[phacops]

This will:

• Add measurements normalization to the span normalization function
• Copy measurements to the segment

[jjbayer]

As discussed, instead of running normalization on spans extracted from events twice, let's make SpanKafkaMessage as strict as the kafka schema and "normalize" the message before producing, like we do here:

relay/relay-server/src/services/processor/span/processing.rs

Lines 398 to 408 in 7970a7c

	sentry_tags.retain(|key, value| match value.value() {
	Some(s) => {
	match key.as_str() {
	"group" => {
	// Only allow up to 16-char hex strings in group.
	s.len() <= 16 && s.chars().all(|c| c.is_ascii_hexdigit())
	}
	"status_code" => s.parse::<u16>().is_ok(),
	_ => true,
	}
	}

We should probably also make the numeric values in SpanKafkaMessage the new FiniteF64, because serde_json will serialize infinity / NaN as null as well.

[jjbayer]

Please see comments before merging. No blockers, but we still cannot be 100% sure that this won't generate new validation errors on the consumer side. To prevent that, I think we need to make SpanKafkaMessage more strict (this would sacrifice some performance) and / or make the consumer more lenient. See also Sentry's development philosophy:

“be conservative in what you send, be liberal in what you accept”.

The second bullet point in the PR description can be removed I believe.

[jjbayer]

See comments about RawValue.