Strange behaviour when using Session.call_module().

**Description of the problem**

Hi all. I'm having issues related to sessions, and the order in which seemingly unrelated commands are issued affecting the final plot in strange ways. I think this is related to #1242 and #733.

Essentially what I'm trying to do is set the dimensions of an inset figure to match the dimensions of the inset map. Perhaps there is a better way to do this, but the way I've found is to use GMT's `mapproject` to calculate the height given the width. PyGMT doesn't expose `mapproject` so I have to use `Session.call_module()`, see the code below.

The problem is that doing this results in strange behaviour, dependent on where exactly the `mapproject` call is issued. I have found a way to arrange the code to get the figure I want, but it is very fragile, and the inset options do not work correctly.

**Full code that generated the error**

This is the "working" sample:

```python
import pygmt
from pygmt.clib import Session
from pygmt.helpers import GMTTempFile

def mapproject_height(region, projection):
with GMTTempFile() as tmp:
args = f"-R{region} -J{projection} -Wh ->{tmp.name}"
with Session() as lib:
lib.call_module("mapproject", args)
h = tmp.read().strip()

return h

region = [-51, -48, -24, -23]
inset_region = [-80, -28, -43, 0]
inset_w = 2

fig = pygmt.Figure()

grid = pygmt.datasets.load_earth_relief(resolution="03s", region=region)
fig.grdimage(grid=grid, projection="M15c", frame=True, cmap="world")

fig.colorbar(frame=["x", "y+lm"], position="JBC+o0c/0.6c+w4.8c")

inset_h = mapproject_height("80W/28W/43S/0S", "M2c")
with fig.inset(position=f"jBL+w{inset_w}c/{inset_h}c+o0.1c", box="+pblack"):
fig.coast(region=inset_region, projection=f"M{inset_w}c",
land="gray", shorelines="thin", water="white")

fig.savefig("test.png")
```

Issues arise when trying to:

- Plot the colorbar after the inset---it is positioned relative to the inset rather than the main figure despite being outside the `with` statement.
- Move the inset to another corner (e.g. TL instead of BL). This results in the inset not being positioned correctly.
- Moving the `mapproject_height` call to the beginning, causes the outer figure to use the region boundaries from the inset for some reason.

Basically this only works if the inset is in the bottom-left, and plotted last, and the `mapproject_height` call is issued right before the `with` statement. This causes problems when I want to make more complex plots.

**Full error message**

```
n/a
```

**System information**

```
PyGMT information:
version: v0.4.1
System information:
python: 3.9.7 (default, Sep 3 2021, 12:37:55) [Clang 12.0.5 (clang-1205.0.22.9)]
executable: /Users/.../.local/share/virtualenvs/bed-mpQI0RL-/bin/python
machine: macOS-11.6-x86_64-i386-64bit
Dependency information:
numpy: 1.21.2
pandas: 1.3.3
xarray: 0.19.0
netCDF4: 1.5.7
packaging: 21.0
ghostscript: 9.53.3
gmt: 6.2.0
GMT library information:
binary dir: /usr/local/Cellar/python@3.9/3.9.7/Frameworks/Python.framework/Versions/3.9/Resources/Python.app/Contents/MacOS
cores: 8
grid layout: rows
library path: /usr/local/Cellar/gmt/6.2.0_1/lib/libgmt.dylib
padding: 2
plugin dir: /usr/local/Cellar/gmt/6.2.0_1/lib/gmt/plugins
share dir: /usr/local/Cellar/gmt/6.2.0_1/share/gmt
version: 6.2.0
```