Overhaul the log records flow

[agagniere]

Stacked on top of #155

Review the log records flow, fix lifetime issues, and simplify allocation and cleaning.

Fix

previously, toReadable duplicated some external strings (body and severity text) but shallow-copied the strings in the attributes, introducing a foot-gun as it appears duplicated but relies on externally-owned data.

Improvement

After reviewing the flow (see .md), it became clear that:

• the SimpleLogProcessor didn't require any duplication at all, as it exports before emit returns
• the BatchLogProcessor requires a full deep copy, then frees all log records when flushing

Furthermore, cleaning requires iterating over all log records, testing for presence of optionals, and, if we were to properly duplicate all strings, iterate over attributes, etc. While in fact, the log records in a batch are all freed at the same time

So this PR:

• introduces asReadable which makes a cheaper shallow copy, without any duplication and any cleaning required, for the SimpleLogProcessor
• makes toReadable a full deep copy
• adds an arena in BatchLogProcessor to allocate the strings of the log records added to the batch, so that cleaning is a single step. Capacity is retained so that the following batches will not required new allocations from the underlying allocator

Misc

Minor improvements are also included:

• use @memcpy instead of a loop in the BatchLogProcessor
• use appendSlice instead of a loop when copying attributes into an ArrayList
• pre-allocated ArrayLists to the right capacity when the number of elements to be inserted is known