⛽ -> ⏲️ Switch from fuel to epoch interruptions.

[itsrainy]

This PR has two main parts:

• Switching to epoch interruptions (the changes to lib/src/execute.rs and lib/src/linking.rs)
• Fixing and refactoring the async_io_methods, which broke as a result of the switch.

The epoch switch is straightforward: set the config value on the wasmtime Engine and then attach a thread to the ExecuteCtx which increments the epoch every 50 µs. We also added a Drop implementation to ExecuteCtx to join that thread.

@jameysharp and I collaborated on chasing down the test issue, namely the async_io_methods test had a race condition (alluded to here) that required it to sleep for 100ms to give the backends time to work. The switch to epoch interruptions made that test flaky, so I added an additional backend and request to the test fixture as a synchronization mechanism between the guest and the server. Once the server has read enough bytes to relieve the backpressure on the write body, it will complete the sync request, signaling to the guest that the write body should be ready.

[itsrainy]

The test failed CI here, so I set it up to run 100 times locally, and it passed 72 times and failed 28 times. So definitely flaky. I'm still unsure why switching to epoch interruptions introduced the flakiness (or made failure more likely), but it seems like the move here is to introduce some backend synchronization mechanism to the test as described here

[cratelyn]

thank you for doing this ✨