Fix inline_string_encode and inline_string_decode

[filip-hejsek]

• Fix incorrect escape decoding, which among other things caused extradata corruption
• Fix Unicode chars being unnecessarily escaped when char is signed
• Move string_codec.cpp to libaegisub so that it can be tested
• Add tests for encoding and decoding
• Remove an unrelated unused include which I noticed while working on this

[arch1t3cht]

Thanks! LGTM except one small nit and the failing Windows CI, which looks like a missing #include <string> in string_codec.h.

[filip-hejsek]

I addressed the nit and added the missing include, let's see if the CI passes now.

Before moving string_codec to libaegisub, it worked without the include, which seems to be because:

• On Linux, #include <string_view> brings in a forward declaration of std::string
• The Winsows CI build has PCH enabled, and agi_pre.h (but not lagi_pre.h) pulls in <string>

Maybe include-what-you-use could help clean up includes in the codebase?

[filip-hejsek]

Oops, I accidentally amended the wrong commit, let me fix that

[arch1t3cht]

Thanks!

Maybe include-what-you-use could help clean up includes in the codebase?

Yes, definitely. There's a bunch of static analysis tooling that I'd like to set up eventually but haven't gotten to yet.

It might also (or additionally) be an option to just also disable PCHs on the Windows and Mac CI. This wasn't done originally in #303 because the Windows CI already takes a very long time, but I since realized that the vast majority of time is spent compiling dependencies rather than Aegisub itself. So disabling PCHs for Aegisub would not have that much of an impact on the CI runtime.