Add client.SetMultipartBoundaryFunc and port Blink/WebKit/Firefox implementations

[rosahaj]

Problem

req currently relies on Go's mime/multipart package to generate unique boundary delimiters for multipart/form-data requests. The boundary delimiters generated by mime/multipart are made up of exactly 60 randomly-generated characters (numbers from 0-9 and lower-case letters from a-f only), e.g. b26dc610159ba676929ced239b36e90370976ad0fc20a615e29e799608d2 (see implementation).

All modern browsers use boundary delimiter formats different from those used by mime/multipart:

• Blink-based browsers (Chrome, Chromium, Edge, etc.) and WebKit-based browsers (Safari, etc.) generate boundary delimiters in the format of ----WebKitFormBoundary[a-zA-Z0-9]{16}.
• Firefox-based browsers generate boundary delimiters in the format of -------------------------\d{1,10}\d{1,10}\d{1,10}.

It's very easy for bot detection tools to detect whether the boundary delimiter used in a multipart/form-data request could have originated from the browser indicated by the request's User-Agent header, with a near-zero risk of false positives, and there's evidence that some websites have implemented such checks (e.g. Discord).

Solution

This PR introduces a new client.SetMultipartBoundaryFunc(fn func() string) method, which accepts a custom function that's called every time a new multipart/form-data request is sent.

Furthermore, it adds ports of the Blink/WebKit and Firefox implementations for boundary delimiter generation to client.ImpersonateChrome(), client.ImpersonateSafari() and client.ImpersonateFirefox().

It also includes test cases for client.SetMultipartBoundaryFunc and the Blink/WebKit and Firefox ports.

[imroc]

Nice feature, thanks for your contribution!