Check for headers correctly in download

[oxinabox]

There was a problem that @mschauer found with

download("https://www.dropbox.com/s/uy6j8wru1nkqhw5/coal.csv?dl=1")

returning the wrong filename.
It was coming back with "coal.csv?dl=1",
but the correct name according to headers is "coal.csv".

I took a look in the headers and realized that they were casing all in lowercase.
Which is allowed, but the download function was doing a case sensitive check on the list of headers,
rather than using the header function on the response.

This PR fixes that.

Could add a test based on the above, but seems a bit fragile.

[oxinabox]

On the way past, I added a thing to break early if the stream eof as it seems that sometimes (all the time?) at first time the stream iofunction is called, the stream is not actually ready,
which means it fails to read the filename from the header, so it defaults the file name,
then makes an empty file there, as it has no body to write to it.

In the cases where the filename is part of the URL this doesn't matter as the original blank file is overwritten. But when it has to be found from the header, it means you get an extra incorrectly named empty file left behind

[mschauer]

Can confirm that this solves the problems with Dropbox links. 👍

[oxinabox]

Julia 1.0 tests failing on linux because
JuliaLang/MbedTLS.jl#193

[mschauer]

In the meanwhile I added your branch as dependency, https://github.com/mschauer/PointProcessInference.jl/blob/master/Manifest.toml

But somehow it still can go wrong, as if DataDeps does use its own dependency

For example, when running

pkg> add https://github.com/mschauer/PointProcessInference.jl

julia> observations, parameters, λinfo = PointProcessInference.loadexample("coal")

I got the old behaviour.

[fredrikekre]

PointProcessInference.jl/Manifest.toml is not used when you add a package, so you will simply get the released version of HTTP.

[mschauer]

Thank you I was not aware of that. To be of help, I went through the code and found no issues (I also checked that the calls to header do the right thing when the key is not found.)

[oxinabox]

Til this is merged you can always get back on this branch by making sure it is added in your current active envirioment.

[mschauer]

Just keeping this on your screen, as you suggested, oxinabox.

[oxinabox]

@quinnj @samoconnor can we merge this?
The CI failure is not related (I think it might be the MBEDTLS problem)