0.21.0

• (breaking-change) The compiler now reuses previous macro run compilations so {{ run(...) }} is only re-run if the code changes
• (breaking-change) Spec: assert { ... } is now it { ... } (thanks @TheLonelyGhost)
• (breaking-change) Renamed Set#merge! to Set#concat
• (breaking-change) Zlib was split into Flate, Gzip and Zlib (bda40f)
• (breaking-change) Crypto::MD5 is now Digest::MD5
• (breaking-change) String#chop is now String#rchop
• (breaking-change) String#to_slice now returns a read-only Slice
• (breaking-change) String can now hold invalid UTF-8 byte sequences, and they produce a unicode replacement character when traversed
• (breaking-change) Removed String#lchomp. Use String#lchop
• (breaking-change) Octal escapes inside strings incorrectly produced a codepoint value instead of a byte value
• (breaking-change) Removed octal escape from char literals
• Fixed compiler performance regression related to cached files (f69e37e)
• Added \xHH escape sequence in string literals
• Char::Reader can now traverse a string backwards
• Enum#to_s now uses pipes instead of commas for flag enums
• IO#read_string is now encoding-aware
• OAuth2::Client now sends application/json Accept header, and considers the expires_in access token property as optional
• Slice can now be read-only
• TCPServer no longer set SO_REUSEPORT to true by default
• Added HTTP::Multipart and HTTP::FormData (thanks @RX14)
• Added File::Stat#pipe?
• Added File.utime
• Added IO#peek
• Added String#strip(arg), String#lstrip(arg), String#rstrip(arg)
• Added String#lchop, String#lchop(prefix), String#rchop and String#rchop(suffix)
• Added String#hexbytes and String#hexbytes?
• Added String#scrub and String#valid_encoding?
• Added includes? macro method for StringLiteral, SymbolLiteral and MacroId (thanks @karlseguin)
• Added "view source" links to GitLab (thanks @ezrast)
• Updated CONTRIBUTING.md guidelines
• Some bug fixes