Function dynamic_element_count_from_bytes 

pub fn dynamic_element_count_from_bytes(
    element: &DataType,
    byte_len: usize,
) -> Option<u32>

Element count a runtime-sized buffer takes from the bytes the caller supplied.

A buffer declared without .with_count(n) has count == 0, which the IR defines as runtime-sized: its element count is whatever the caller passed for it on this dispatch. This function is the ONE rule for deriving that count, byte_len divided by the element width, sub-byte element types included.

The CPU reference, CUDA, and WGPU all size runtime-sized writable buffers through this rule, so a countless declaration reads back the same element count on every backend. A second copy of this arithmetic is how the three paths drifted into returning 4 words, 1 word, and 0 words for one program.

Returns None when the element type has no fixed width or the count does not fit u32.