Implement encoding for more types

[LPGhatguy]

This PR is based on #11. This PR enables users to write ktx2 images using the crate without having to manually encode core data structures. It still requires a lot of code even with this PR, so I'd like to follow it up with another PR that includes a higher-level API without sacrificing performance or simplicity.

I extended pseudo_enum to enable configurable integer sizes. I also hid the inner value in favor of exposing a value method as it makes consuming code a bit cleaner. This was necessary because many of these types are narrower than 32 bits.

I made the LENGTH associated constant public on a couple of types and implemented as_bytes for them. For DataFormatDescriptorHeader::as_bytes, I added an argument of the DFD block size because it's not part of the type. I think it maybe should be part of the type.

I changed some members that have invariants about being non-zero from their primitive types to NonZero* types. In addition, I changed the relevant cases doing arithmetic on those values to handle incorrect cases to prevent panics or silent incorrect values.

I put the one existing test into a mod test block as is convention and started adding round-trip tests for encoding and decoding. This helped uncover a couple bugs as I was working on the implementation. I think that this should use property-based testing if we can, but I have no experience with it in Rust.

Here is a GitHub gist containing the code to write a single layer RGBA8 sRGB image using this crate after this PR: https://gist.github.com/LPGhatguy/a42ccf5bed9c7a099d2ee7a169d03c80

This shows that there is still a lot of nontrivial code required to encode a KTX2 image. I am interested in exploring APIs to make this much easier and I think it's important to do that in this crate.

Checklist

• cargo clippy reports no issues
• cargo doc reports no issues
• cargo deny issues have been fixed or added to deny.toml
• cargo test shows all tests passing
• human-readable change descriptions added to the changelog under the "Unreleased" heading.
  • If the change does not affect the user (or is a process change), preface the change with "Internal:"
  • Add credit to yourself for each change: Added new functionality @githubname.

Description

Related Issues

[Ralith]

Is NonZeroU8 here useful enough to justify the ergonomic overhead compared to a plain u8?

[LPGhatguy]

I think it's worthwhile. The value must be non-zero or we can't encode the struct, and I think it's helpful to keep as_bytes non-fallible.

[Ralith]

Same ergonomic question as above.

[Ralith]

I think that this should use property-based testing if we can, but I have no experience with it in Rust.

I've gotten some good results from some initial attempts to apply proptest to quinn's delay queue.

[LPGhatguy]

Review feedback should be addressed.

[Ralith]

Still ambivalent about the NonZero stuff, but not inclined to block on it.

[LPGhatguy]

Hey! Not blocked on this, but anything else I can do to help move this towards merging?

[expenses]

Hi! #21 is going to make encoding the DFDs in combination with this a bit easier. I'm pretty happy with the changes here and it'd make things a lot easier to have these merged before #21.