[TIR] Add a new intrinsic count leading zeros for LLVM and SPIR-V

[masahi]

This adds a new intrinsic clz to LLVM and SPIRV per discussion in #7669 (comment).

The motivation is to compute integer ceil(log2(x)) without using log2, since SPIR-V doesn't support log2 on fp64. Using this new intrinsic, we can compute integer ceil(log2(x)) as follows:

def ceil_log2(x):
    clz = tvm.tir.clz(x)
    bits = int(x.dtype[-2:])
    ceil_log2 = tvm.tir.if_then_else(x & (x - 1) == 0, bits - clz - 1, bits - clz)
    return ceil_log2

LLVM has ctlz intrinsic which we can use directly. SPIRV doesn't have clz intrin but it does have FindUMsb.
https://llvm.org/docs/LangRef.html#llvm-ctlz-intrinsic
https://www.khronos.org/registry/spir-v/specs/1.0/GLSL.std.450.html

Some discussion points:

• The result is defined only for positive integers.
• For now I didn't support clz on zero, because the LLVM doc above suggests clz(0) is not well supported by backends. See is_zero_undef argument in the llvm doc for more details.
• Do we want to support clz on negative integers?
• What about floating point?

@tqchen @junrushao1994 @vinx13 @zxybazh

[masahi]

Good to go? Or any comments on the discussion points above @tqchen @junrushao1994 @vinx13 @zxybazh

[junrushao]

LGTM :-)