Drop workarounds from BitConverter

[EgorBo]

It seems like #11413 is already fixed so we don't need these workarounds. It's hard to trace down what exactly fixed that, a lot of struct/RA/cast-morphing work have been done since then.

double Foo(ulong x) => BitConverter.UInt64BitsToDouble(x);

codegen (before and after this change):

; Method Pp:Foo(long):double:this
G_M16062_IG01:
       vzeroupper 
G_M16062_IG02:
       vmovd    xmm0, rdx
G_M16062_IG03:
       ret      
; Total bytes of code: 9

Also, remove [MethodImpl(MethodImplOptions.AggressiveInlining)] All of these methods are less than 10 bytes of IL and always pass "below always_inline" for both RyuJIT and Mono.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-numerics
See info in area-owners.md if you want to be subscribed.

Issue Details

---

It seems like #11413 is already fixed so we don't need these workarounds. It's hard to trace down what exactly fixed that, a lot of struct/RA work have been done since then.

Also, remove [MethodImpl(MethodImplOptions.AggressiveInlining)] all of these method are less than 10 bytes of IL and always pass "below always_inline" for both RyuJIT and Mono.

Author:	EgorBo
Assignees:	-
Labels:	area-System.Numerics
Milestone:	-

[EgorBo]

😄 looks like I've already tried to drop these workarounds and there is still a CQ issue when these casts are used in e.g. conditions #42348 (comment)