Use caching to improve performance

[konstin]

dataclasses-json is very slow when parsing lists of objects because it introspects the dataclasses every time. This pull request caches the slowest of those calls using python's built-in lru_cache, Reducing the decoding time for my test json from 21s to 2.1s.

There are commits: With 56c3c9d the already failing test keeps failing, while the remainder of the tests passes. With cbe4d87 pytest fails on collecting tests cases due to some type being unhashable.

[lidatong]

Hey there, thanks for raising this and the contribution, I appreicate it a lot. I definitely agree that dataclasses-json is doing a rather inefficient thing by traversing the schema on every load.

Coincidentally, I previously considered addressing this with @lru_cache, but as you mention there is the hashable input limitation. This can be solved by using a more feature-rich cache implementation (e.g. cachetools). Specifically, cachetools supports a key function that allows you to specify how to derive a hashable key from your object.

Please feel free to implement, or I'm also happy to take this over the line. Thanks again!

[konstin]

Unfortunately I don't have the time to look into cachetools.

[lidatong]

Superseded by #175