The primary goal here is to ensure we have a consistent heap allocator with reasonably predictable performance across all platforms.

The secondary goals are:

Statically link the full tcmalloc profiler.
Update and clean up the cpu profiler bindings (e.g. add flush).
Build system should use tcmalloc-and-profiler by default, but provide options for the system allocator and tcmalloc-minimal.
Write a playbook to help with data collection from the field, possibly additional shell helpers.

This is a fairly open-ended task; suggestions welcome.