Use large-model addressing for calls when in non-PIC mode.

The main use for non-PIC code at present is JIT code, and JIT code can
live anywhere in memory and reference other symbols defined anywhere in
memory, so it needs to use the "large" code model.

func_addr and globalsym_addr instructions were already using `movabs`
to support arbitrary 64-bit addresses, so this just makes calls be
legalized to support arbitrary 64-bit addresses also.

cranelift/filetests/isa/intel/binary64.cton

@@ -473,8 +473,11 @@ ebb0:
     ; asm: movzbq %dl, %rsi
     [-,%rsi]             v351 = bint.i64 v301   ; bin: 0f b6 f2
 
-    ; asm: call foo
-    call fn0()                                  ; bin: e8 PCRel4(%foo-4) 00000000
+    ; TODO: x86-64 can't encode a direct call to an arbitrary 64-bit address in
+    ; a single instruction. When we add a concept of colocated definitions, this
+    ; test can be re-enabled.
+    ; disabled: asm: call foo
+    ; disabled: call fn0()                                  ; bin: e8 PCRel4(%foo-4) 00000000
 
     ; asm: movabsq $0, %rcx
     [-,%rcx]            v400 = func_addr.i64 fn0        ; bin: 48 b9 Abs8(%foo) 0000000000000000

cranelift/filetests/isa/intel/legalize-libcall.cton

@@ -2,6 +2,7 @@ test legalizer
 
 ; Pre-SSE 4.1, we need to use runtime library calls for floating point rounding operations.
 set is_64bit
+set is_pic
 isa intel
 
 function %floor(f32) -> f32 {

cranelift/filetests/isa/intel/prologue-epilogue.cton

@@ -1,6 +1,7 @@
 test compile
 set is_64bit
 set is_compressed
+set is_pic
 isa intel haswell
 
 ; An empty function.