Fix crashes in mgd77magref with a numeric -D

[Esteban82]

Claude analyzed mgd77magref and found a few bugs around the Dst buffer that lead to crashes. This PR proposes the fix. The scripts below show them (on current master the first three segfault; the last one segfaults after the error for the bad segment).

1. -D<number> segfaults: parse() wrote the value into CM4_D.dst before it was allocated (it is allocated later in the module body).
2. -D-8 behaved as -D8: the leading - was skipped before atof.
3. With several input records a numeric -D read past the end of its 1-element dst array.
4. A segment outside the Dst time span freed dst in MGD77_cm4field, so the next single-record segment wrote into a NULL pointer.
   dst is now allocated at parse time and freed only in Free_Ctrl. Adds test/mgd77/magref_dst.sh.

Tested with:

export GMT_END_SHOW=off
echo "-30 45 0 2000.08700533" | gmt mgd77magref -A+y -Fxyz/3 -D8
echo "-30 45 0 2000.08700533" | gmt mgd77magref -A+y -Fxyz/3 -D-8
printf -- "-30 45 0 2000.087\n-30 45 0 2000.5\n-30 45 0 2001.2\n" | gmt mgd77magref -A+y -Fxyz/3 -D8
printf -- "> bad\n-30 45 0 1900.5\n-30 45 0 1900.6\n> good\n-30 45 0 2000.5\n" | gmt mgd77magref -A+y -Fxyz/3

Written with Claude Sonnet 5.5, reviewed with Claude Opus 5.5.

This is the output with this fix:

-8.91517189718  3.78735619674   10.5974519577
-15.5064972641  4.44507960625   19.2142882413
-8.88607779405  3.77783244797   10.6115537859
-6.01975730977  2.8274257026    17.2528952866
-8.80171128582  2.58243039681   12.9876225555
> bad
INTDST -- Error: T (15202; 15203) LIES OUTSIDE OF DST TABLE TIME SPAN [35839; 60675] -- ABORT
mgd77magref [ERROR]: this segment has a record generating an error.
Unfortunately, this means all other eventually good
records are also ignored. Fix the bad record and rerun the command.
> good
-3.84802225651  2.23032262522   14.4369055462