Document simulation time

[m-mueller678]

There is not much documentation on how simulated time works (and the runtime in general). I have gathered these facts from the source code and would like to open a PR with all of these documented. I wanted to ask if they are correct before doing that.

• A task can be spawned either via madsim::task, madsim::runtime::NodeHandle::spawn, or madsim::runtime::Runtime::block_on.
• tasks spawned via madsim::runtime::Runtime::block_on all have the same node ID, and it is distinct from node IDs belonging to an actual NodeHandle.
• tasks spawned via madsim::task belong to the same node from which madsim::task was used. Using it outside the runtime panics.
• Each task belongs to exactly one node (possibly the block_on one)
• All nodes share a common clock
• within a task, std::time functions return simulation time. This is achieved by overriding functions usually provided by libc.
• The runtime is single threaded, all task polling happens within the block_on call.
• If the future passed to block_on completes while there are still nodes running, those nodes remain in the runtime. They may continue to make progress during a future call to block_on.
• Tasks that are ready are polled in a pseudo-random order
• Time never advances during the polling of a task
• Time advances by a small pseudo-random amount after each time a task is polled (currently 50-100 ns)
• If no tasks are ready, but some tasks are sleeping, time advances until the earliest sleep deadline plus some small amount (currently 50ns)