Avoid corrupting data when calling seekend

[nhz2]

Fixes #109

I changed the default seekstart to error when applied to a stream in :write mode, and reset the stream if it is in :read mode. It also calls seekstart on the wrapped IO.

seekend(s) should be equivalent to skip(s, big number) in which case it should currently error if the stream is not in :read mode.

Also changed skip to return the stream to be more consistent.

[nhz2]

@Marlin-Na Does this solve your issue? @quinnj @bicycle1885 This is ready for review.

The current behavior of seekend on a stream in :write mode can very easily lead to corrupt data.

For example currently:

julia> stream = GzipCompressorStream(open("foo.txt.gz"; write=true))
TranscodingStreams.TranscodingStream{GzipCompressor, IOStream}(<mode=idle>)

julia> write(stream, collect(1:10000))
80000

julia> seekend(stream)
TranscodingStreams.TranscodingStream{GzipCompressor, IOStream}(<mode=write>)

julia> write(stream, collect(10001:20000))
80000

julia> close(stream)

julia> reinterpret(Int,transcode(GzipDecompressor,read("foo.txt.gz")))
ERROR: zlib error: invalid block type (code: -3)

With this PR the error happens while writing, but no data is lost

julia> stream = GzipCompressorStream(open("foo.txt.gz"; write=true))
TranscodingStreams.TranscodingStream{GzipCompressor, IOStream}(<mode=idle>)

julia> write(stream, collect(1:10000))
80000

julia> seekend(stream)
ERROR: ArgumentError: not in read mode
Stacktrace:
 [1] seekend(stream::TranscodingStreams.TranscodingStream{GzipCompressor, IOStream})
   @ TranscodingStreams ~/juliadev/TranscodingStreams/src/stream.jl:308
 [2] top-level scope
   @ REPL[49]:1

julia> write(stream, collect(10001:20000))
80000

julia> close(stream)

julia> reinterpret(Int,transcode(GzipDecompressor,read("foo.txt.gz")))
20000-element reinterpret(Int64, ::Vector{UInt8}):
     1
     2
     3
     4
     5
     6
     ⋮
 19996
 19997
 19998
 19999
 20000

[jakobnissen]

I don't think it's a good idea to read the entire file here. Normally, when a user uses seek, it's specifically to avoid processing a whole file.
I understand that it may not be possible to seek into a compressed file and get the position of the decompressed stream, but I believe in that case, seekend should simply not be implemented, else we risk frustrating users when an operation they believe to be constant time suddenly takes ages.
We can then provide the functionality of reading the whole file under a function with another name.

[nhz2]

Yes, I agree that it is much better for Base.seekend(stream::TranscodingStream) not to be implemented than to silently do the wrong thing. I'm not sure how to go about removing this method though. I created a PR JuliaIO/CodecZlib.jl#68 to remove its use in CodecZlib.jl's tests.

[nhz2]

I can split this PR into three parts, one for each change, to make reviewing easier.