fix: make clean Uint8Array

[rvagg]

See https://github.com/browserify/commonjs-assert/blob/v1.5.0/assert.js#L293-L294 for an example of why this can fail, the slice doesn't remove from the backing buffer. Yes that comparison is wrong but it's the version of assert that we're stuck with in Webpack without explicitly pulling in a newer version.

I'm also considering if maybe we shouldn't be doing a Uint8Array.from(o) in the coerce() function just to be sure.

[rvagg]

fwiw I can't get some CID-related tests to pass in the browser because of this, they pass in Node.js but not in the browser and I'm not sure whether it's to do with borc preferring Uint8Array's in the browser but handling Buffers on Node.js (might even come from an fs call), or the comparison algorithm differences in assert that I get in the browser vs Node.js. Either way, it's going to bite other people.

[vmx]

CI is failing weirdly and still running on Node,js 10.

[rvagg]

Yeah, CI should be ignored for now, this is still using the .travis.yml from js-ipld-dag-cbor, I imagine Mikeal will pull in his standard Github Actions workflows which do test and publish. This PR is against his WIP branch which isn't quite ready for merging into master, although with this patch it's working for me in js-datastore-car.

[mikeal]

tests are passing but coverage is still not 100% which is where it needs to be before i turn on the automated releases.