Reader now reads headers into stream, and some checks are added there
and in UploadFileClientHandler, so if headers are too long, they are
treated as bad request.
Now Client reads the first headers block, then LocalWebserver decides
which Handler to use. In case of "/upload", UploadFileHandler is used.
But now it only knows the "path" parameter. If that's valid, actual
UploadFileClientHandler is created, which reads the contents of the
request and, when finds there an "upload_file" field, starts saving it
in the directory specified by "path".
With that we don't need temp files approach from Reader class.
Instead of keeping them in memory. Temp file name is generated to point
into ScummVM's working directory. That means that if user wanted to
upload file to the place with sufficient size, it would instead be
uploaded to ScummVM's working directory. Yet it's too early to parse the
target directory, so there is no way to generate temp file name within
that directory.
That means that if current buffer is over, reader will stop reading and
it's safe to call readResponse() again, so it would continue from the
place it left.
That should be part of the Client, I guess. Reader is not ready to
continue reading from place it stopped, but it already works as it
should for the case when whole content is available.
