switch to meson build

This is a work in progress effort to modernize our build machinery and use meson, recently adopted by the scipy stack.

In doing this I also follow the approach adopted by the scipy stack of using Tempita, Cython's built-in templating engine.

Rough plan:

• Change the build infrastructure to use meson-python and pyproject.toml
• Implement environment-dependent search for mumps (right now I target conda-forge) Library names can now be passed via a meson build option.
• Figure out the development workflow (running tests etc)
• Update CI
• Explore conda packaging

Also closes #19 (closed) and #11 (closed) (I cleaned up a bunch of code for a good measure).

README.md

@@ -2,21 +2,20 @@
 
 Python bindings for the [MUMPS](http://mumps-solver.org/): a parallel sparse direct solver.
 
-
-# Installation
+## Installation
 
 `mumpy` works with Python 3.10 and higher on Linux, Windows and Mac.
 
-The recommended way to install `mumpy` is using [`conda`](https://conda.io/):
+The recommended way to install `mumpy` is using `mamba`/`conda`.
+
 ```bash
-conda install -c conda-forge mumpy
+mamba install -c conda-forge mumpy
 ```
 
 `mumpy` can also be installed from PyPI, however this is a more involved procedure
 that requires separately installing the MUMPS library and a C compiler.
 
-
-# Usage example
+## Usage example
 
 The following example shows how mumpy can be used to implement sparse diagonalization
 with Scipy.
@@ -46,3 +45,19 @@ def sparse_diag(matrix, k, sigma, **kwargs):
     opinv = LuInv(matrix - sigma * identity(matrix.shape[0]))
     return sla.eigsh(matrix, k, sigma=sigma, OPinv=opinv, **kwargs)
 ```
+
+## Development
+
+`mumpy` recommends [Spin](https://github.com/scientific-python/spin/). Get spin with:
+
+```bash
+pip install spin
+```
+
+Then to build, test and install `mumpy`:
+
+```bash
+spin build
+spin test -- --lf  # (Pytest arguments go after --)
+spin install
+```