load_earth_mask: Keep data's encoding to correctly infer data's registration and gtype information

[seisman]

Description of proposed changes

See #2629 for the bug report.

The bug is because:

1. In the GMT accessor, we rely on grid.encoding["source"] to determine the grid's registration and gtype
2. For the earth_mask dataset, we use return grid.astype("int8"). It copies the old xarray object to a new one, converts the new grid's data array to int8 type and returns the new xarary object. (ref: https://docs.xarray.dev/en/stable/generated/xarray.DataArray.astype.html)
3. During the "int8" conversion, the grid.encoding information is lost.

This PR fixes the bug by directly working on the grid's data array. I also updated the test to check if the returned grid has the correct gtype.

Closes #2629.

Reminders

• Run make format and make check to make sure the code follows the style guide.
• Add tests for new features or tests that would have caught the bug that you're fixing.
• Add new public functions/methods/classes to doc/api/index.rst.
• Write detailed docstrings for all functions/methods.
• If wrapping a new module, open a 'Wrap new GMT module' issue and submit reasonably-sized PRs.
• If adding new functionality, add an example to docstrings or tutorials.
• Use underscores (not hyphens) in names of Python files and directories.

Slash Commands

You can write slash commands (/command) in the first line of a comment to perform
specific operations. Supported slash commands are:

• /format: automatically format and lint the code
• /test-gmt-dev: run full tests on the latest GMT development version

[seisman]

BTW, the upstream discussions in pydata/xarray#1297 and pydata/xarray#1614 provide more background information.

[weiji14]

Looks good, just one minor suggested check on the dtype for test_earth_mask_01d_with_region