feat: add sent reason for late arriving spans

[VinozzZ]

Which problem is this PR solving?

• Please provide a more descriptive reason for late-arriving spans #835

Short description of the changes

This PR adds a new cache for send_reasons that Refinery produces for a given configuration.
The new cache acts as a mapping between the string representation of send reason and a uint.
The cache is used to reduce the memory footprint of the trace cache.

I made the design based on the assumption that the send reason has a low cardinality for a given configuration.
We decided not to clear out the cache on config reload so that we can avoid needing a lock on read operation.

fix #835

[kentquirk]

I just want to note that we went over this on a call and the benchmarks showed that this cache design is extremely fast, since it's lockless on read. We also tested for race conditions in parallel operation.