feat(s2n-quic-core): add Cached clock implementation

[camshaft]

Description of changes:

This adds a mechanism to cache the clock value after the first call. This can reduce the number of times we read from the system clock, where we don't care about high precision between events.

Call-outs:

I've also added a timer debugger that you can query all of the locations in the timer tree that print where they are and if they're armed or not.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[WesleyRosenblum]

this cache doesn't need a TTL?

[camshaft]

Since this borrows an inner clock for 'a, the lifetime of the borrow is the TTL, if that makes sense. Its intended use is something like this:

{
    let clock = clock.cached();

    for packet in ack_packets {
        packet.ack_time = clock.get_time();
    }
}

{
    let clock = clock.cached();

    for packet in lost_packets {
        packet.lost_time = clock.get_time();
    }
}

In this example, we'll only query the cache at most twice, regardless of how many packets are in the two iterators.

[camshaft]

We've got a few ad-hoc examples of doing this in the codebase than I'm going to clean up after this is merged:

s2n-quic/quic/s2n-quic-transport/src/endpoint/mod.rs

Lines 103 to 112 in e2f32e5

	let mut now: Option<Timestamp> = None;
	queue.for_each(|mut header, payload| {
	let timestamp = match now {
	Some(time) => time,
	None => {
	now = Some(clock.get_time());
	now.expect("value should be set")
	}
	};

[WesleyRosenblum]

ah nice. Would it be possible to a add a debug_assertion or something to prevent misuse? Something like if the difference between the cache and now is > x milliseconds, panic?

[camshaft]

the problem is what would the x be without triggering false positives... :/

[WesleyRosenblum]

I'm ok to merge this though for now, though if its mainly used in pretty tight loops I think we could start with something like 1 or 2 ms

[camshaft]

I think that's way too low, especially in debug mode. I was thinking 1s+. But yeah we can figure something else later. I'm not super worried about it being misused since it borrows a clock rather than owning it so it'd really only live as long as a stack frame.