[Breaking] Remove the type piracy of several Base functions

[DilumAluthge]

This pull request removes the type piracy of several Base functions. Specifically, it removes the following pirated methods:

1. Base.startwrite(io::Any)
2. Base.closewrite(io::Any)
3. Base.startread(io::Any)
4. Base.closeread(io::Any)

According to #745 (comment), this is a breaking change. Therefore, I took advantage of the opportunity to bump the version number to 1.0.0.

For context, see: #745 (review) and JuliaLang/julia#42080.

[DilumAluthge]

cc: @fredrikekre @KristofferC @quinnj

[DilumAluthge]

@quinnj I'll probably need your help to fix the tests.

[c42f]

None of startwrite, startread, closeread are defined in Base yet and IIUC there's no particular plan to add them. (I could potentially see closeread being added for symmetry with closewrite in the future. Not so sure about startread/startwrite.)

From my point of view I like @quinnj's current solution better for closewrite, somewhat hacky as it may be. The point is that closewrite and the new Base.closewrite should have exactly the same semantics because they solve the same problem. Thus, they deserve to share method tables.

SemVer is a blunt tool as usual... but if we need to release a breaking change anyway, I'd suggest we figure out how to unify things rather than separate them.

[KristofferC]

The point is that closewrite and the new Base.closewrite should have exactly the same semantics because they solve the same problem. Thus, they deserve to share method tables.

Why should closewrite(io) = nothing exist? And in that case, it should be moved to Base.

[c42f]

Why should closewrite(io) = nothing exist? And in that case, it should be moved to Base.

Well I agree you're right about that! If this is a nonsensical method, we should just remove this fallback method from HTTP.jl and declare a breaking change.

What I disagree about is something subtler: this PR makes HTTP.closewrite and Base.closewrite into separate generic functions with disjoint method tables and clashing exported names. I don't think we should do this because they have the same semantics.