std: Expose get_mut() on I/O wrappers

[alexcrichton]

Currently, it is impossible to reset the timeout on a TCP socket inside of a
buffered writer or a buffered stream. This is because the method requires &mut self, but the buffered objects only provide shared access to the contained
object.

By allowing mutable access, the possibility for corruption of the stream is
exposed. The tradeoff is that many on-the-fly configuration methods are now
possible without having to unwrap the buffered streams.

[brson]

I was hitting a similar type-system situation recently where I couldn't call tell while using iterators because I had lost access to the stream.

I wonder if I/O is meant for an interior-mutability world, or if there are some yet-undiscovered Rusty abstractions that would be better for I/O.

Mutating inner I/O objects scares me: is there any chance this could lead to memory unsafety?

[alexcrichton]

This may also mean that some of our signatures are wrong. Perhaps seek should and the property setting flavors should all take &self instead of &mut, only leaving read and write to to taking &mut. Although seeking on a buffered reader could corrupt the buffered state if it didn't know about it...

[aturon]

cc me

[alexcrichton]

I talked with @aturon tonight at great length about this, and we came to the conclusion that this is likely not the right approach. I will need to think some more before submitting another attempt at this.