Lift commonly used arm64 floating point instructions `scvtf` and `sxtl`

[plafosse]

What is the feature you'd like to have?
scvtf and sxtl aren't lifted and should be. This binary also exibits this issue: #5134

struct_array_test_arm64.zip

[galenbwill]

scvtf has four variants:

1. scalar, integer: arm64: lift scvtf #5459
2. scalar, fixed-point
3. vector, integer: intrinsic only
4. vector, fixed-point
   Since we don't support fixed-point types, we can only lift the integer variants. Without first-class vector support, we can only fully lift the scalar variants -- vector variants will have to use intrinsics.

Update: the relevant PR was subsumed and superseded in 12d4783 and is on dev release 4.1.5383.

[galenbwill]

Update on sxtl, etc.: this led me to lifting a raft of related/similar instructions:

1. sshll (which sxtl is an alias of)
2. uxtl (which is partially lifted) and usshll (which is not, despite being what uxtl is an alias of)
3. sshl, sshr (which are left- and right-shifting non-immediate analogs to sshll)
4. ushl, ushr (which are the unsigned variants of sshl and sshr)

...and this led me back to my old dilemma, the question of whether to "lift by unrolling" or leave the intrinsics when it comes to vector operations. (Unlike scvtf, these are solely vector operations, so the "only implementing the scalar variants" approach is a non-starter... and the answer to this dilemma will affect the lifting of scvtf.)

[galenbwill]

This issue should include:

• scvtf
• ucvtf
• sxtl{2}/sshll{2}
• uxtl{2}/ushll{2}
• sshl, sshr
• ushl, ushr