Fix to no copy when format with yansi_term

[zzau13]

This PR eliminates the need to reallocate while formatting.
For this I have forked the ansi_term library to refactor to fmt::Display allowing to introduce flow with a closure that takes the current formatter as argument.
It will depend on yansi-term instead of ansi_term, which I will keep.

[zbraniecki]

@botika what's your take on yansi-term vs termcolor? @brendanzab says that termcolor may be a better long term bet.

I'm happy to land this PR, but I'm also curious on your opinion on that :)

[zzau13]

The library is very simple, just enter a prefix and a suffix. So it does not need much maintenance nor can it innovate significantly. With the fork in the ansi_term I have left the minimum to fulfill its function as quickly as possible

The reason for the PR is to remove all unnecessary mallocs in the formatter.

[zbraniecki]

I understand. I'm just curious on your take on long-term replacing ansi_term or yansi_term with termcolor?

[zzau13]

It is more optimal implement fmt::Display than io::Write. It forces you to use his buffer or stream output. I think the only thing this library should do is put a prefix and a suffix around the format flow, and let the user use whatever writer they want. So you couldn't use fmt::Formatter for output stream. And I am unaware of compatibility outside of unix systems

[zzau13]

I've been looking at termcolor in more depth, just as I expected, it's forced to reallocate, here allocate to stack and here allocate to writer (wherever the writer writes).

So, the main problems of termcolor are unnecessary use of memory and lack of abstraction of the writer

[zzau13]

I updated the write bench, the results between io::Write and fmt::Display are:

Teams io::Write time: [423.37 ns 430.62 ns 436.27 ns] Teams fmt::Write time: [291.71 ns 292.90 ns 294.35 ns] 

• implementation io::Write
• implementation fmt::Display

The results are clear. The fastest way to write in Rust is to implement fmt::Display.

[zbraniecki]

Thank you!