Update fuzz tests to released crowbar API

[talex5]

In #237 (comment), @hannesm noted that some cstruct functions can be used to access memory outside of the original cstruct (but inside the original buffer), which seems a bit surprising. I noticed that the fuzz tests actually check for the safer behaviour, so I tried running them to see why they were passing. It turns out that they have bit-rotted and no longer compile.

This PR updates the tests to use the released crowbar API. I ran afl-fuzz on fuzz.exe, after a few minutes it reported three failures:

sub 0 204
sub: FAIL

When given the input:

    [#1 [44943; 0; 0]; 0; 204]
    
the test failed:

    check false

That is:

utop # let whole = Cstruct.create 44943 in
       let empty = Cstruct.sub whole 0 0 in
       Cstruct.sub empty 0 204;;
- : t = {buffer = <abstr>; off = 0; len = 204}

add_len: FAIL

When given the input:

    [#1 [60159; 8; 0]; 32768]
    
the test failed:

    check false

set_len: FAIL

When given the input:

    [#1 [61056; 0; 0]; 512]
    
the test failed:

    check false

I'm not sure what we expect the correct behaviour to be here, but either the tests or the code should be changed.

[hannesm]

NB: in mirage/mirage-tcpip#394 i removed various set_len calls. IMHO set_len and add_len should be removed, and sub be fixed. thanks!

[dinosaure]

Same as @hannesm, sub should be fixed. However, we need to see if mirage-tcpip relies on this behavior.

[avsm]

thanks!