export profiling as http handler

[zdyj3170101136]

Is your feature request related to a problem? Please describe.
when rss increase, but python heap profiling overhead is too large that i could not enabled it in production.

Describe the solution you'd like
accept profiling as http handler, return heap profiling data.
GET /debug/pprof/heap

so i could reduce overhead by only trigger profiling when rss is very high or only sampling 10% percent of my machine.

[sarahec]

Would #781 help your use case? Then you could configure a web server with whatever custom trigger you need and have it issue a redirect to show the result.