HTTP libraries in general are tough because HTTP in practice requires understanding of at least DNS and TLS and in a lot of cases L4 networking as well. If you try to target application developers you end up with leaky abstractions (unread H/1 body draining behavior when you don't even know what HTTP version you're using) or a library low level enough that application developers don't know how to use it. See rust's hyper and reqwest as an example of an attempt at that split.

As someone who once attempted implementing a http library, I can assure you that the protocol itself is at least equally tough as being aware of all the TLS / L4 edge cases.

I’d argue that implementing both TCP and IP from scratch is easier than implementing an HTTP library.