Add `ConvolveScanner` class and subclasses (replace `new_convolve_scanner` and `new_double_scanner` function factories)

Something like this:

class SingleConvolveScanner:
    def __init__(self, array: ArrayLike, transform: Callable):
        self.array = array
        self.transform = transform

    def __call__(self, history_subset: ArrayLike, multiplier: float) -> tuple[ArrayLike, float]:
        new_val = self.transform(multiplier * jnp.dot(self.array, history_subset))
        latest = jnp.hstack([history_subset[1:], new_val])
        return latest, new_val

Pro

• Easier to inspect
• Explicitly calling an object constructor makes it even clearer we're making a new thing.

Con

• A bit more verbose
• Probably don't want to encourage mutating scanners in place e.g.:

my_scanner = SingleConvolveScanner(jnp.array[0.5, 0.3, 0.2]), transform=lambda x: x)
my_scanner.array = jnp.array([0.7, 0.2, 0.1])

Locate new_convolve_scanner and new_double_scanner and trace their call sites first. Use those call sites to determine the needed ConvolveScanner subclasses, then verify the factories are replaced without changing scanner behavior; no test or file path is named in the issue.