Improve the two-step grid reading when source is a matrix

[PaulWessel]

Description of proposed changes

See #3408 for the initial problem. Here is what goes wrong:

1. We first read the container header from a file. The file is a memory reference to a matrix. This returns a grid structure with a header.
2. Then grdgradient is called with the same file to make an intensity grid. This all works.
3. Later, we wish to read the data from the matrix source. However, despite me implementing more meaningful memory file names like @GMTAPI@-S-I-G-M-G-N-000000, the old code has not been updated: It loops over the registered objects and finds the first one that matches the specs (image, input, etc). This turns out to be the grid header from (1) above instead of item 0. So we go down the rabbit hole and do strange things.

Time to implement a new scheme that extracts the information directly from the filename.

Closes #3408.

[PaulWessel]

I have made a few changes. First, I am parsing the full name and get the various items. For now I am only checking direction and return the object ID. I also made a small change that when we check for the validity of a given object I ignore if it has been used before (it has). I added @seisman example C code as well and two tests (with and without shading) which both passes. All other tests pass as well. There are more simplifications that can be done in gmt_api.c but I want to do that very carefully...
I also fixed a memory leak from grdimage.

[PaulWessel]

Closes #3408.