In the case where we have multiple keys stored in the same bucket,
a next search for the second (or later) key leads to a bogus lookup.
This occurs as the key is not memoized between consecutive calls
of the next function, and so conflicts with some state information
about the LUT. This commit resolves this issue, and improves the unit
tests to specifically look for this case.

missing still: reject non-aks elems!
brush N_structelem
fix print on fundec

Carefully parse multiple expressions in return or multi-op with-loop

Closes #2252

See merge request !134

Install config.h file

See merge request !137

We do make use of config.h in some sac.h included headers, and not
having this leads to errors when compiling sac programs.

Change to centos8 and ubi18

See merge request !136

ubi14 and centos6 are now EOL, repos have been de-activated.

Hotfix for several bugs

Closes #2255

See merge request !135

In the long run, we want to change that type into something bigger. However, currently, the backend uses a N_num to encode this which, internally uses int....
We cannot even make it unsigned as we use -1 to indicate lack of ravel knowledge!

tests for the other bugs would have been nice; unfortunately,  they are rather big and trying to minimise them was not possible....

we now can AKD withloops with empty results as well :-)

functions would only be "opened up" for improving the inferred return type if they were still instances of a wrapper function. Isolated instances that had been dispatched and their wrapper functions deleted would never be improved wrt their return type again.
This could lead to a situation where the inferred type was AKD but the official return type was AUDGZ/AUD.
This incompatability then led to descriptor mismatches in the generated code which led to uninitialised dim/shape info.
Erratic runtime behaviour was the consequence!

noinline
int[.] offset2idx (int offset, int[.] shp)
{
    res = [];
    for( i=2; _ge_SxS_( i, 0); i=_sub_SxS_(i, 1)) {
        res = _cat_VxV_ (offset , res);                         // offset should be in brackets but the TC would previously choke on it :-)
    }
    return res;
}

int main()
{
   vect = offset2idx ( 8, [3,4,3]);

   return _sel_VxA_([0], vect);
}

in particular WLBSC needs to be run directly before WLTRA! added another run in phase wlt....

adjusted cmake / sac2crc config for OSX BigSur / and found a better solution...

See merge request !132

adjusted cmake / sac2crc config for OSX BigSur / and found a better solution for the longlong problem in tag_executionmode.c

Eradicate old types

See merge request !131