Optimize reverse proxy

Do some profiling and get the latency as low as possible, possible replace Go's reverse proxy with https://github.com/valyala/fasthttp, test unix domain sockets in Lambda as well, check heap and see what can be pooled. It's also making some assumptions about having concurrency of 1 at the moment, since that's the case with Lambda.

Make sure it's not decompressing gzip and recompressing either, not sure what the proxy does but I know net/http's client does this transparently, test with some larger responses.

- [ ] profile cpu
- [ ] profile mem
- [ ] test larger responses
- [ ] implement any quick wins
- [x] either remove Accept-Encoding or ignore when already compressed