feat: Exponential read-ahead cache

[kylebarron]

Replace simple prefetch buffer with exponential readahead cache for more efficient metadata fetching, and improve Python error handling.

Supersedes and closes #83, closes #89

Change list

• ReadaheadMetadataCache grows fetch sizes (configurable via initial_size and multiplier parameters)
• Caches sequential blocks from file start to minimize redundant network requests
• Default configuration: 32 KiB initial fetch with 2x multiplier for subsequent fetches
• Updates Python bindings
• Adds minimal Python error handling via PyAsyncTiffError enum, so the existing test that checked for FileNotFoundError would still pass. Better exception handling is tracked in Improved Python error handling #144

[weiji14]

Recommend to add a unit test 🙂

[weiji14]

The slice here might return an empty chunk, and the .push would fail with "IOError(Error { kind: UnexpectedEof, message: "failed to fill whole buffer" })". Maybe re-check the logic here?

[kylebarron]

Can you think of an example of a unit test that would catch this?

[kylebarron]

I added some unit tests in 00a6e72 (#140)