Add failing idempotence test

[TomerAberbach]

I noticed that camelCase is not idempotent, which I think is a bug. i.e. if camelCase is meant to return a camel-cased version of the input, then camel-casing it again should not result in a diff because it's already been camel-cased.

I've added a failing property-based test for this, but haven't figured out what the implementation bug is exactly. Feel free to merge as-is, but I may update this PR with an actual bug fix if I have time.

An example of a case that fails is "{ A":

• camelCase("{ A") -> "{A"
• camelCase(camelCase("{ A")) -> "{a"

[sindresorhus]

Thanks for the detailed investigation and property-based test!

After analyzing this, the non-idempotent behavior is an intentional trade-off. The function preserves special characters like @, #, :: in the output (see tests for mGridCol6@md, A::a, etc.). Fixing idempotence would require treating these as separators, which would break existing behavior.

The edge case you found ("{ A" → "{A}" → "{a}") is extremely rare in practice, and the function is designed to convert to camelCase rather than be a no-op on already-camelCased input.

I've documented this behavior in the README to avoid confusion. Closing as won't fix.