List patterns: Add initial support for enumerables

[alrz]

This deviates from the draft spec in a few ways, in particular:

• Splitting TryGetElement into a pair of HasElementAt and GetElementFrom{Start,End} methods: To facilitate subsumption checking, a new test node is added to track the count. It narrows down a virtual temp by count > index at each element access. This also correctly eliminates forthcoming element tests where the result is known:

  • HasElementAt(i) implies HasElementAt(j) where i >= j
  • !HasElementAt(i) implies !HasElementAt(j) where i <= j

  The "relation condition" is inserted as a constant value test just like indexables, because count > 1 && !(count > 2) produced by a fixed-length pattern (see below) leads to a single remaining value. This test won’t reach codegen.

  It’s still possible to lower a sequence of (element test, element evaluation) as TryGetElement for a slightly smaller code size.

• Removal of Last(): To test the end of the sequence, a negated test for the next element is used instead: !HasElementAt(N)

• Removal of Count(): Trailing patterns are prefixed by HasElementAt(N-2) to ensure there’s sufficient elements before calling GetElementFromEnd for each subpattern after the slice.

Note: this only covers IEnumerable<T> at the moment.

[alrz]

This is ready for an initial review pass @jcouv @dotnet/roslyn-compiler
Had some questions, added as PROTOTYPE comments in the code.
Thanks.

(Some changes are pulled from #66481 and #66692)

[jaredpar]

Thanks for contributing this as always @alrz 😄

At the moment we are a bit overbooked for work. I think the earliest we'll be able to book this for review is this summer. It is on our list of potentials though.

Sorry, this cycle has been a bit more resource constrained than anticipated.

[ChadNedzlek]

Curious about the state of this, as it's been a couple years, and this issue is referenced in the regex group.