[RUM v3] Optimize fields

[jalvz]

This includes:

• Derive parent_id, trace_id and transaction_id in the span from its transaction

• Make required booleans optional, defining absence as false ( sync)

• Make start and duration int.

• Remove unneeded fields: transaction.marks.navigationTiming, http.url

[axw]

Make start and duration int.

They're currently "number", which allows the agent to send an integer as well as a float. What's the reason for making the schema stricter?

[jalvz]

They're currently "number", which allows the agent to send an integer

Good point, I haven't noticed.
But while looking at how to implement this, I changed my mind about the approach: it's probably simpler for us that the agent embeds the spans inside their transactions.

There is no easy way that we can take fields out of a transaction decoder and pass them as arguments to the spans decoder.

I think we would need an specific HandlerRawModel for this, or a flag inside the processor, and it would break the uniformity of the work flow (decoder signatures, etc).

wdyt?

[axw]

If the RUM agent will always have spans within transactions, then I think we wouldn't need or want a separate span decoder. The RUM transaction decoder would return a Transformable which internally combines info from the transaction and embedded span objects to produce the transaction and span docs.

[jalvz]

All boolean fields are optional, so they can default to false when missing.

I think that with #3659 this can be closed.