ICO: Add header validation in strict mode

[RunDevelopment]

This PR adds more validation for the ICO header in strict mode.

Changes:

• The reserved first two bytes are now checked to be zero.
• The type is now checked to be either ICO or CUR.

[197g]

Appreciate the idea, more validation to keep strict mode from changing behavior silently on reserved data 👍

[197g]

I'd use split_first_chunk here to instead of an IO-facing library. Alternatively, but only since this is uniformly u16, you might turn header into [[u8; 2]; 3] and read it with as_flattened? Lots of alternatives that do not involve an IO error for a statically sized array and constant condition.

[RunDevelopment]

TBH, I just copied what TGA is doing. Should we not do that there too?

I also considered the u16 approach, but didn't like it because it makes it really easy to mess up endian. E.g.

let mut header = [0_u16; 3];
reader.read_exact(bytemuck::cast_slice_mut(&mut header))?;
let [reserved, id_type, entries] = header; // only correct on LE systems

But I agree with you that the IO interface is a little overkill. Maybe we could just have a 2 or 3 util methods like this:

// Panics when OOB
fn read_u16<Endian>(slice: &[u8], offset: usize) -> u16;

I saw this pattern in Chrome, so that might be an idea.

[RunDevelopment]

Appreciate the idea, more validation to keep strict mode from changing behavior silently on reserved data 👍

That, and I want to lay the groundwork for supporting CUR files eventually.