ShaderBuffer changes to reduce allocation and change set_data to take an iterator

[beicause]

Objective

Some unnecessary allocations about ShaderBuffer:

1. ShaderBuffer::set_data allocates a new Vec each time and doesn't reuse existing data.
2. ShaderBuffer::set_data takes T, which usually causes unnecessary buffer.set_data(iter.collect::<Vec<_>>()).
3. ShaderBuffer::from(T) allocates a Vec then ShaderBuffer::new allocates a duplicated Vec. In addition, if T is Vec, the ShaderBuffer::from can't reuse T.

Solution

1. ShaderBuffer::new now accepts Vec<u8> instead of &[u8].
2. ShaderBuffer::from now accepts Vec<T> where T: bytemuck::NoUninit. From<T: ShaderType + WriteInto> is replaced by ShaderBuffer::from_value.
3. ShaderBuffer::set_data now accepts impl Iterator<Item = T> instead of a single T value.

Note that previous if vec is empty buffer.set_data(vec) will insert at least one zeroed element. I reserved this behavior by peeking iterator and extending zeros.

Testing

cargo r --example fps_overlay --features bevy_dev_tools and cargo r --example storage_buffer

[beicause]

Thinking again, I perfer to remove this behavior. Automatically inserting zeros is a bit surprising. I'd rather let the caller handle empty iterators themselves.

[beicause]

Note: #24938 is an alternative and conflicts with this.