Add top-level compile time to @time report (alternative)

[IanButterworth]

@NHDaly, in thinking about #39760 I wondered if this might be an alternative way to achieve what #37938 is doing?

Opening as a draft given I'm not sure whether it's better!

It only reports top-level compilation time if it exceeds 0.5 seconds, and it's intentionally low-precision, given higher precision would seem odd given the 0.5 second cutoff.
If users are trying to time with @time and care about top-level compilation < 0.5 seconds they probably want to be using @btime anyway.

julia> @eval function f(a)  # Expensive function to compile. (Thanks @yuyichao!)
                  $(quote $([:(a = a + a * 10 + (a > 0 ? $i : 0.2)) for i in 1:401]...) end)
                  $(quote $([:(a = a + a * 10 + (a > 0 ? $i : 0.2)) for i in 1:401]...) end)
                  return a
              end
f (generic function with 1 method)

julia> @time f(2)
  0.000000 seconds + 3.0 seconds of top-level compilation
-Inf

julia> x = 1
1

julia> @time f(x)
 19.168382 seconds (527.44 k allocations: 19.444 MiB, 100.00% compilation time)
Inf

julia> @time f(x)
  0.000015 seconds (1 allocation: 16 bytes)
Inf

julia> for i in 1:4
       @time sleep(1)
       end
  1.002893 seconds (4 allocations: 128 bytes)
  1.004821 seconds (4 allocations: 128 bytes)
  1.005353 seconds (4 allocations: 128 bytes)
  1.004838 seconds (4 allocations: 128 bytes)

julia> Threads.@threads for i in 1:4
       @time sleep(1)
       end
  1.005255 seconds (42 allocations: 2.438 KiB)
  1.005194 seconds (4 allocations: 128 bytes)
  1.005236 seconds (30 allocations: 2.125 KiB)
  1.005329 seconds (83 allocations: 5.562 KiB)

If moved forward, will need working into other timing macros

[JeffBezanson]

Macro expansion should not have side effects, so this is not a great way to implement this.

[IanButterworth]

Closing as deemed to be a bad approach, and the @time @eval trick added to the docs in 1.6 is a helpful tool for investigating this issue