[m3msg] Use slice-backed buffers in messageWriter

[vdarulis]

What this PR does / why we need it:

Using multiple slices as a backing store allows us to keep the same semantics, have less pointer-chasing and reduced time in critical section, as we no longer have to hold a lock and prevent new writes into the queue while a scan is in progress, in exchange for occasional batch allocs and more copying. This change keeps the same semantics, prioritizing new writes.

Special notes for your reviewer:

Does this PR introduce a user-facing and/or backwards incompatible change?:

Does this PR require updating code package or user-facing documentation?: