TypeMap: Cease trying to guarantee sorting

[vtjnash]

We've found it can be hard to maintain specificity ordering linearly of method dispatch (c.f. #30171), this instead just embraces that and explicitly doesn't care and just adds an extra list to keep track of any inconsistencies. It turns out that seems to look a lot like our existing ambiguity analysis code! Plus, it'll hopefully mean we can make some easy speculative optimizations to the TypeMap which will hopefully make it faster to load code (#15048, #21760, #16418).

Closes #32747
Closes #29594

[JeffBezanson]

Would be good to get some before and after package loading times.

[vtjnash]

We don't expect this to make much difference (as seen below), since it does roughly the same amount of work, and just does it more correctly. It does however gives us more tools to do so in the future, as well as prepares us to fix a threading race condition.

on master:

$ ./julia -q
julia> @time using DataFrames
[ Info: Recompiling stale cache file /home/vtjnash/.julia/compiled/v1.3/DataFrames/AR9oZ.ji for DataFrames [a93c6f00-e57d-5684-b7b6-d8193f3e46c0]
 17.232868 seconds (2.40 M allocations: 133.970 MiB, 0.18% gc time)

$ ./julia -q
julia> @time using DataFrames
  0.879511 seconds (1.30 M allocations: 81.718 MiB)

on PR:

$ ./julia -q
julia> @time using DataFrames
[ Info: Recompiling stale cache file /home/vtjnash/.julia/compiled/v1.3/DataFrames/AR9oZ.ji for DataFrames [a93c6f00-e57d-5684-b7b6-d8193f3e46c0]
 16.930647 seconds (2.43 M allocations: 136.471 MiB, 0.47% gc time)

$ ./julia -q
julia> @time using DataFrames
  0.897301 seconds (1.38 M allocations: 86.582 MiB)

[JeffBezanson]

Ok, that's great. I didn't expect a significant change, but good to sanity check.