Fix heap-buffer-overflow in Array deserialization

[silverfoxy]

This PR fixes a bug where Array::from_vec and Array::From disagreed on capacity calculation, causing heap-buffer-overflow when deserializing with certain serializers (e.g., bincode)
I added simple regression test for the capacity mismatch issue.

The Bug
from_vec used cap = arr.len() while From assumed cap = len.next_power_of_two().

When deserializing 100 elements:

1. Deserializer creates Vec with capacity 100
2. from_vec stores pointer with capacity 100
3. Later, From reconstructs assuming capacity 128
4. Writing to indices 100-127 overflows the heap buffer
   This affects serializers like bincode that create exact-capacity Vecs. serde_json accidentally masked the bug because push() doubles capacity.

The Fix
from_vec now resizes the Vec to len.next_power_of_two() before forgetting it, ensuring the allocation matches what From expects.

[silverfoxy]

This makes clippy happy.

  --> src/estimator.rs:98:34
   |
98 |     pub(crate) fn representation(&self) -> Representation<P, W> {
   |                                  ^^^^^     ^^^^^^^^^^^^^^^^^^^^ the same lifetime is hidden here
   |                                  |
   |                                  the lifetime is elided here
   |
   = help: the same lifetime is referred to in inconsistent ways, making the signature confusing
   = note: `-D mismatched-lifetime-syntaxes` implied by `-D warnings`
   = help: to override `-D warnings` add `#[allow(mismatched_lifetime_syntaxes)]`
help: use `'_` for type paths
   |
98 |     pub(crate) fn representation(&self) -> Representation<'_, P, W> {