Problem:

The current writer implementation is backed by Java heap arrays. While mostly sufficient for writing small test recordings, it becomes a bottleneck when we try writing out test recordings containing 10s of 1000s of events for eg. parser performance benchmarks

Proposed Solution:

Implement an opt-in memory-mapped file support in addition to heap backed support.

Add support for fixed-sized chunks we can flush frequently, even for heap array based writers, thus alleviating the need to keep the full recording description in memory until it is finalized.