Performance: reserve body to avoid frequent reallocations and copies

[ArnaudBienner]

When doing some profiling on a client application doing a lot of downloads, I realized my application spent ~13% of its time doing std::string::append (compiled with gcc and -03, even more when compiling in debug mode with -O1: ~30%).

I realized this could be improved by reserving the output buffer (res.body) since we know the output size if Content-Length header is present.

[ArnaudBienner]

I've implemented an alternative in #1782.

[yhirose]

@ArnaudBienner I don't think std::string::append gets called after your code, because the rest of code after the line 6069 in Server::write_response_core is only for writing res.body or data from a content provider to a network socket. So res.body won't be changed. Could you confirm that your code really works? Thanks!

[ArnaudBienner]

Sorry: I made that work on an older version of the library, and it looks like the patch wasn't correctly imported with git apply.

I made a new PR by copying the code by hand, to the correct location this time: #1788.

I'll close this PR.