Use only strings in xarray.Dataset metadata (fatiando#25)

The `harmonica.load_icgem_gdf` function populated the `Dataset.attrs`
with metadata derived from the file header. Two of these fields
(`attributes` and `attributes_units`) were lists. This was fine while
using xarray with netcdf4 but it does not work if trying to use scipy to
save/load the `Dataset`. And since xarray 0.11 stopped installing
netcdf4 automatically, we can't load our sample gravity and topography
data without adding netcdf4 as a dependency.

Instead of doing that, `load_icgem_gdf` now converts the metadata to
strings. New copies of the two sample grids have been generated using
scipy (netcdf3).

Author: leouieda

data/etopo1-0.5deg.nc.xz

@@ -1,2 +1,2 @@
-etopo1-0.5deg.nc.xz bb44b4aaa49778159dc3655ae0c0c23ad88789328829557ba7ea0113f26baa7f
-gravity-earth-0.5deg.nc.xz 84340b771ac91695b0874639a1497b6b81e0f5a8d21ac51cc221d1c48598f5a7
+etopo1-0.5deg.nc.xz 9f21f7c946e649389dce28631de94fa05217750e9ebc77df4311458c33f0682b
+gravity-earth-0.5deg.nc.xz 02c62a251225e2e76722a80d87c488160fbdcae2c1a8bbc2386c32e68ea8f43a

data/gravity-earth-0.5deg.nc.xz

@@ -42,7 +42,9 @@ def fetch_gravity_earth():
         (``height_over_ell``). Coordinates are latitude and longitude.
 
     """
-    return _load_xz_compressed_grid(POOCH.fetch("gravity-earth-0.5deg.nc.xz"))
+    fname = POOCH.fetch("gravity-earth-0.5deg.nc.xz")
+    data = _load_xz_compressed_grid(fname, engine="scipy")
+    return data
 
 
 def fetch_topography_earth():
@@ -65,7 +67,9 @@ def fetch_topography_earth():
         The topography grid (in meters). Coordinates are latitude and longitude.
 
     """
-    return _load_xz_compressed_grid(POOCH.fetch("etopo1-0.5deg.nc.xz"))
+    fname = POOCH.fetch("etopo1-0.5deg.nc.xz")
+    data = _load_xz_compressed_grid(fname, engine="scipy")
+    return data
 
 
 def _load_xz_compressed_grid(fname, **kwargs):

harmonica/datasets/registry.txt

@@ -49,6 +49,10 @@ def load_icgem_gdf(fname, **kwargs):
     if "height_over_ell" in metadata:
         height = float(metadata["height_over_ell"].split()[0])
         data_vars["height_over_ell"] = (dims, np.full(shape, height))
+    # Can't have lists in the Dataset metadata to make it compatible with netCDF3. This
+    # way the data can be saved using only scipy as a dependency instead of netcdf4.
+    metadata["attributes"] = " ".join(metadata["attributes"])
+    metadata["attributes_units"] = " ".join(metadata["attributes_units"])
     grid = xr.Dataset(data_vars, coords=coords, attrs=metadata)
     # Check area from header equals to area from data in cols
     area = [