[mono] Optimize boxing in Pattern Matching idioms

[EgorBo]

This PR optimizes boxing for pattern matching idioms. Basically, it copies existing optimizations from CoreCLR to mono, see impBoxPatternMatch (one of them was recently added by me for coreclr, see #1817)

1) box + isinst + brtrue/brfalse --> ldc.i4.0/1 + brtrue/brfalse

public static int Case1<T>(T t)
{
    if (t is int)
        return 42;
    return 0;
}
/*
    IL_0000: ldarg.0
    IL_0001: box !!T
    IL_0006: isinst [System.Private.CoreLib]System.Int32
    IL_000b: brfalse.s IL_0010
    IL_000d: ldc.i4.s 42
    IL_000f: ret
    IL_0010: ldc.i4.0
    IL_0011: ret
*/

2) box + isinst + ldnull + ceq/cgt.un --> ldc.i4.0/1

public static bool Case2<T>(T t)
{
    return t is int;
}
/*
    IL_0000: ldarg.0
    IL_0001: box !!T
    IL_0006: isinst [System.Private.CoreLib]System.Int32
    IL_000b: ldnull
    IL_000c: cgt.un
    IL_000e: ret
*/

3) box + isinst + unbox.any --> nop

public static int Case3_1<T>(T o)
{
    if (o is int x)
        return x;
    return 0;
}

// or

public static int Case3_2<T>(T o) => o is int n ? n : 42;

// or

public static int Case3_3<T>(T o)
{
    return o switch
    {
        int n => n,
        string str => str.Length,
        _ => 0
    };
}

Tests can be found here: https://github.com/dotnet/runtime/tree/master/src/coreclr/tests/src/JIT/Generics/Conversions/Boxing

Sharplab.io link to observe IL code for the samples above.

Codegen diff example:

static bool Test(int n) => Validate(n);

static bool Validate<T>(T t) => t is int;

Before (for Test):

0000000000000000	pushq	%r14
0000000000000002	pushq	%rbx
0000000000000003	pushq	%rax
0000000000000004	movl	%edi, %ebx
0000000000000006	movabsq	$0x7fdf9c5090c0, %rax
0000000000000010	movabsq	$0x7fdfad0268e8, %r14
000000000000001a	leaq	0x1be800(%r14), %rdi
0000000000000021	movl	$0x14, %esi
0000000000000026	callq	*(%rax)
0000000000000028	movl	%ebx, 0x10(%rax)
000000000000002b	movq	(%rax), %rax
000000000000002e	cmpq	%r14, (%rax)
0000000000000031	sete	%al
0000000000000034	addq	$0x8, %rsp
0000000000000038	popq	%rbx
0000000000000039	popq	%r14
000000000000003b	retq

After:

0000000000000000	movb	$0x1, %al
0000000000000002	retq