Fix lossly cast in value_to_usize

[davidpdrsn]

Checklist

• I have read the Contributor Guide
• I have read and agree to the Code of Conduct
• I have added a description of my changes and why I'd like them included in the section below

Description of Changes

Turns out going key_path::get(1) infers 1 to be an i32, since get takes an impl Into<Value>. So we somehow need to cast from at least i32 to usize for accessing lists and arrays with key paths.

To remove the lossly conversion I can think of two options:

1. Use TryFrom and panic if the conversion fails. This would only happen when accessing a very large list.
2. key_path::get is also used for accessing maps (hence why it takes impl Into<Value>) but we could make separate functions for list/array and map access. Perhaps get_seq(usize) and get_map(impl Into<Value>).

I personally prefer option one because I like how list/array and map access is the same, but it is perhaps a bit disingenuous. I like how it matches rusts' Index trait though. What do you think @bnjbvr?

Related Issues

Fixes #76

[bnjbvr]

Since mirror-mirror is a lib, it's nice to avoid panicking as much as possible, especially if considering fuzzing situations where randomly generated keypaths would hit such a case trivially. Could we instead return None for those cases, and add in the doc-comment for KeyPath and the macro that indexes larger than usize::max_value() won't work?

[bnjbvr]

(Sorry I didn't see the question at first.)

From options 1 and 2, option 2 looks slightly nicer to me, for the reason that it doesn't panic.

Returning None as I suggested could also work nicely, in my opinion: if you pass an index type that's bigger than a usize, you couldn't feed it to Vec::get anyways, so it conceptually returns nothing, hence None. The panic also conceptually works, but that morphs what would be a compile-time error (indexing with a larger-than-usize) into a runtime-error, and that would still be hit easily with fuzzers.