math: remove x86 modf asm

the int part was wrong when -1 < x <= -0 (+0.0 instead of -0.0)
and the size and performace gain of the asm version was negligible

src/math/i386/modf.s

-.global modf
-.type modf,@function
-modf:
-	mov 12(%esp),%ecx
-	fldl 4(%esp)
-	fxam
-	fnstsw %ax
-	sahf
-        jnp 1f
-        jc 2f
-1:      fld1
-        fld %st(1)
-1:      fprem
-        fnstsw %ax
-        sahf
-        jp 1b
-        fstp %st(1)
-        fsubr %st(0),%st(1)
-        fxch %st(1)
-        fstpl (%ecx)
-        ret
-
-2:	fstpl (%ecx)
-	mov 4(%ecx),%eax
-	and $0x80000000,%eax
-	mov %eax,4(%esp)
-	flds 4(%esp)
-	ret

src/math/i386/modff.s

-.global modff
-.type modff,@function
-modff:
-	mov 8(%esp),%ecx
-	flds 4(%esp)
-	fxam
-	fnstsw %ax
-	sahf
-        jnp 1f
-        jc 2f
-1:      fld1
-        fld %st(1)
-1:      fprem
-        fnstsw %ax
-        sahf
-        jp 1b
-        fstp %st(1)
-        fsubr %st(0),%st(1)
-        fxch %st(1)
-        fstps (%ecx)
-        ret
-
-2:	fstps (%ecx)
-	mov (%ecx),%eax
-	and $0x80000000,%eax
-	mov %eax,4(%esp)
-	flds 4(%esp)
-	ret

src/math/i386/modfl.s

-.global modfl
-.type modfl,@function
-modfl:
-	mov 16(%esp),%ecx
-	fldt 4(%esp)
-	fxam
-	fnstsw %ax
-	sahf
-        jnp 1f
-        jc 2f
-1:      fld1
-        fld %st(1)
-1:      fprem
-        fnstsw %ax
-        sahf
-        jp 1b
-        fstp %st(1)
-        fsubr %st(0),%st(1)
-        fxch %st(1)
-        fstpt (%ecx)
-        ret
-
-2:	fstpt (%ecx)
-	mov 6(%ecx),%eax
-	and $0x80000000,%eax
-	mov %eax,4(%esp)
-	flds 4(%esp)
-	ret

src/math/x86_64/modfl.s

-.global modfl
-.type modfl,@function
-modfl:
-	fldt 8(%rsp)
-	fxam
-	fnstsw %ax
-	sahf
-        jnp 1f
-        jc 2f
-1:      fld1
-        fld %st(1)
-1:      fprem
-        fnstsw %ax
-        sahf
-        jp 1b
-        fstp %st(1)
-        fsubr %st(0),%st(1)
-        fxch %st(1)
-        fstpt (%rdi)
-        ret
-
-2:	fstpt (%rdi)
-	mov 6(%rdi),%eax
-	and $0x80000000,%eax
-	mov %eax,8(%rsp)
-	flds 8(%rsp)
-	ret