Add timeout support

[jethrogb]

The following operations in std require some form of timeout support:

• TcpStream::connect_timeout
• TcpStream::set_read_timeout
• TcpStream::set_write_timeout
• thread::sleep
• thread::sleep_ms
• thread::park_timeout
• thread::park_timeout_ms
• Condvar::wait_timeout
• mpsc::Receiver::recv_timeout

Currently, the first just doesn't use a timeout, and the rest errors/panics.

The ABI currently does not support waiting a specified amount (only no wait or indefinite wait): https://docs.rs/fortanix-sgx-abi/0.1.0-rc1/fortanix_sgx_abi/struct.Usercalls.html#method.wait . It's also not clear if timeouts make sense in SGX setting, in particular: is it the case that users depend on accurate timeouts for security purposes?

[jethrogb]

sleep, park, and Condvar can be fixed by allowing more values for the timeout parameter to the wait usercall.

More work is needed to support read/write/connect. Here are some ideas:

Option 1

Add a way to cancel async usercalls, then do the async call in conjunction with wait with timeout. If wait returns first, cancel the async call.

Cancellation here means that the call should return immediately. It may still return succesfully if it was completed in the mean time.

Option 2

Add *const Usercall/*mut Return to wait(), to specify timeout for a specific call.

Concern: Is this a reasonable API? How does this behavior interact with event_mask and allowed spurious wakeups?

Option 3

Add timeout() usercall, very similar to “Option 2”, except that there is a dedicated API.

Concern: API size increase.

[jethrogb]

Should the API use deadlines or timeouts?