Matrix grid doesn't always work with different central meridians

**Description of the problem**

The issue was first reported in https://github.com/GenericMappingTools/pygmt/issues/390, but was thought to be fixed by GMT PRs #3813 and #3829. The GMT master branch works for some projection parameters but doesn't always work.

**Full script that generated the error**

```python
import pygmt
fig = pygmt.Figure()

fig.grdimage("@earth_relief_01d_g", projection="Q123/0/15c", cmap="geo")

fig.shift_origin(yshift="8c")

dem = pygmt.datasets.load_earth_relief(resolution="01d", registration="gridline")
fig.grdimage(dem, projection="Q123/0/15c", cmap="geo")
fig.savefig("map.pdf")
```

**Output**


**Image difference**

In the PyGMT, we test several different projection parameters. Here are the ones that are known to fail:

- `Q123/0/15c`
- `Q123/30/15c`
- `Q180/0/15c`
- `Q180/30/15c`

Here are the ones that are known to work:

- `Q0/0/15c`
- `Q0/30/15c`

The same parameters work well for General Stereographic projection (i.e., changing **Q** to **S**).

Note: Those tests were originally written by @weiji14.