Support for reading GeoKeyDirectory

[gschulze]

This PR implements support for reading the GeoKeyDirectory.

The structure is based on https://github.com/developmentseed/aiocogeo-rs/blob/main/src/geo_key_directory.rs, thanks kylebarron for the hint!

@weiji14 I split this off the coordinate transformation branch, as this is a prerequisite for debugging stuff conveniently in QGIS.

[weiji14]

Thanks @gschulze! Just some docstring and minor code suggestions/questions for now.

[weiji14]

Could use a ref here instead, and change all the *count to count above?

Suggested change

	count: u16,
	offset: u16,
	) -> TiffResult<u16> {
	if location_tag.is_some() {
	return Err(TiffError::FormatError(TiffFormatError::Format(format!(
	"Key `{key_tag:?}` did not have the expected value type."
	))));
	}
	if count != 1 {
	count: &u16,
	offset: u16,
	) -> TiffResult<u16> {
	if location_tag.is_some() {
	return Err(TiffError::FormatError(TiffFormatError::Format(format!(
	"Key `{key_tag:?}` did not have the expected value type."
	))));
	}
	if count != &1 {

[gschulze]

I can do that, but I'm unsure whether I completely understand the rationale. I think you want me to change it to a reference because of the contract of the equality operator: fn eq(&self, &other: Self): bool. Is there a performance impact to this, or is it just a matter of style? I can imagine the compiler will optimize this anyway. For me, *count == 1 is more intuitive than count == &1, as in my mind, I'm comparing values rather than references. Changing from count: u16 to count: &u16 also consumes more memory on the stack, four times as much on 64-bit systems. Anyway, I'm happy to change it if you want me to; I just wanted to share my thoughts. Is there any best-practices document you follow that I can relate to?

[kylebarron]

I'd love to hear rationale either way or see some resources on this, but for types that implement Copy I think it's simpler to not use a reference in the signature parameter

[weiji14]

Ah ok, I was approaching this more as a matter of style rather than performance. Still learning Rust on the go, and didn't realize that references to u16 would be stored as 64-bit (!!), so happy for you to stick with *count == 1.

[gschulze]

I'm no expert on this topic either, but inspecting the code via Compiler Explorer might reveal more insights.

[gschulze]

@weiji14 Thanks for your comments; I updated the PR and incorporated most of your suggestions. See my comment on the usage of *count == 1 vs count == &1 above, though.

[weiji14]

Thanks again @gschulze, just one small doc suggestion, otherwise feel free to merge!