Instruction after CFU instruction is getting wrong result

Create `proj/hps_accel/src/main.c` with the following contents:

```cc
#include

#include "base.h"
#include "riscv.h"

__attribute__((optimize("align-functions=32"))) int check_post_cfu_instruction(
int input) {
register int output;
int tmp;
asm volatile(
// Some padding. The alignment of the CFU instruction seems to matter.
"nop \n\t"
"nop \n\t"
"nop \n\t"
"nop \n\t"
"nop \n\t"
"nop \n\t"

// Set up an input to our CFU instruction.
"li a6, 0; \n\t"

// These four instructions seem to need to be stores.
"sw %[input],%[tmp] \n\t"
"sw %[input],%[tmp] \n\t"
"sw %[input],%[tmp] \n\t"
"sw %[input],%[tmp] \n\t"

// This can be any instruction.
"nop \n\t"

// Our CFU instruction.
".word ((CUSTOM0) | (regnum_a7 << 7) | (regnum_a6 << 15) | (regnum_a6 << "
"20) | ((0) << 12) | ((5) << 25)) \n\t"

// The value stored into %[output] is wrong.
"mv %[output],%[input] \n\t"

// Changing the following nop to repeat the line above, or swapping it
// with the line above "fixes" the bug.
"nop \n\t"

"nop \n\t"
"nop \n\t"
: [output] "=r"(output), [tmp] "=m"(tmp)
: [input] "r"(input)
: "a6", "a7"

);
return output;
}

int main(void) {
init_runtime();

int v = check_post_cfu_instruction(42);

// This should print the value 42, but when the bug is occurring, it prints 0.
printf("Value: %d\n", v);

return (0);
}
```

Then run:

```sh
(source environment; cd proj/hps_accel/; make PLATFORM=sim load)
```

Expected output:

```
....
Value: 42
```

Actual output:
```
....
Value: 0
```

Swapping the `mv` instruction with the `nop` that follows it, adjusting the alignment +/- 8 or changing some of the store instructions to `nop`s makes the problem go away.

I'm guessing this is probably a bug in VexRiscv's CFU interface - possibly some bad interaction with instruction pipelining.