Dataset modification in Reader.__attrs_post_init__ is overwritten by base class initialization

[alexandre-borowczyk]

titiler-multidim version: v0.7.1

Description:
I'm working with a ZARR 4D dataset (T, Z, Y, X) where the time dimension T always has length 1 and needs to be squeezed for proper processing. I attempted to modify the dataset after the self.opener call in Reader.__attrs_post_init__, but discovered that modifications are overwritten during initialization.

Note: This makes caching logic also ineffective

Root Cause:
The dataset is subsequently set in the base class titiler.xarray.io.Reader.__attrs_post_init__, which overwrites any modifications made in the multidim-specific Reader subclass before the super().__attrs_post_init__()call.
Proposed Solution:
I've resolved this by modifying titiler.multidim.reader.Reader.__attrs_post_init__:

# Add this block to set self.input before base class initialization
self.input = get_variable(
    self.ds,
    self.variable,
    sel=self.sel,
    method=self.method,
)

# Change from super().__attrs_post_init__() to:
super(Reader, self).__attrs_post_init__()

This ensures dataset modifications persist by setting self.input before the base class processes it, and the explicit super(Reader, self).__attrs_post_init__() call skips the immediate parent's reinitialization.

Question:
I'm happy to submit a pull request with this fix, but as I'm new to titiler, I want to ensure this aligns with the intended initialization workflow. Is this the correct approach for applying dataset transformations in subclasses, or would you prefer a different pattern?