time-with-isolation test is drifting and often confuses people

[saethlin]

On my machine, with a silly for _ in 0..100_000 {}, the isolated clock already runs at about half the speed of the actual clock. That's probably somewhat tolerable, but rust-lang/rust#122420 is going to push us farther away from reality.

We might want to re-calibrate this magic value:

miri/src/clock.rs

Lines 4 to 9 in b228407

	/// When using a virtual clock, this defines how many nanoseconds we pretend are passing for each
	/// basic block.
	/// This number is pretty random, but it has been shown to approximately cause
	/// some sample programs to run within an order of magnitude of real time on desktop CPUs.
	/// (See `tests/pass/shims/time-with-isolation*.rs`.)
	const NANOSECONDS_PER_BASIC_BLOCK: u64 = 5000;

Or if our goal is really just an order of magnitude, we should make the test failure a lot more clear so reviewers/contributors are less confused by it.