fix: correct calculation error when series type is f32

[ssrtw]

Problem Cause

The Talib function expects input of type double. When the input Series has a data type of f32, it leads to abnormal calculation results. Initially, the to_float function of Polars was used for type conversion, but this function does not handle the conversion from f32 to f64 since f32 is already a floating-point type. Passing the chunk pointer of type f32 directly as double to the Talib calculation results in memory alignment issues, affecting the calculation results. As encountered in issue #11, the attached image shows that fields like adjusted_close are of type f32.

Changes Made

1. Added the cast_series_to_f64 function to utils.rs, which casts non-f64 types and rechunks.
2. Replaced the usage of to_float with cast_series_to_f64 in overlap.rs to convert non-f64 Series types, ensuring all input data for Talib functions are of type double.

polars_ta_extension/src/overlap.rs

Line 21 in dcfd75a

let input = &mut inputs[0].to_float()?.rechunk();

let input = &mut cast_series_to_f64(&inputs[0]);

pub fn cast_series_to_f64(series: &Series) -> Series {
    match series.dtype() {
        DataType::Float64 => Ok(series.clone()),
        _ => series.cast(&DataType::Float64),
    }
    .unwrap()
    .rechunk()
}