Test autopush/endpoint w/ low MALLOC_ARENA_MAX values and jemalloc

[pjenvey]

We appear to be suffering from glibc malloc using excessive memory w/ our allocation patterns, similar to the situation described here:

http://codearcana.com/posts/2016/07/11/arena-leak-in-glibc.html

We need to see what memory improvements can be seen w/ autopush/endpoint under a MALLOC_ARENA_MAX=1

We currently have a couple nodes running at a more conservative MALLOC_ARENA_MAX=2. We're still seeing the same rate of growth, however the total RSS used has been cut by about 33% under endpoint, and about 20% under autopush.

[pjenvey]

Also jemalloc may show more improvement over glibc malloc, support for enabling it was added in mozilla-services/autopush#1100

[bbangert]

Memory use is pretty solid now and the crypto is faster. Remaining work is obsolete per Rust move this year.