Fix some corner cases in the custom JSX parser and add tests

[jviide]

PR #38 introduces a new custom parser implementing the complete JSX syntax. This pull request adds tests to some corner cases, and suggests fixes for them:

• JSX allows property names that contain hyphens. This can be fixed by JSON.stringifying the property names.
• Boolean attributes following non-boolean ones got ignored (e.g. <a b="1" c /> yielded the result h('a', { b: '1' }). This can be fixed by switching to MODE_WHITESPACE when we encounter a whitespace (and after calling commit()).
• out.replace(/,\(\{(.*?)$/, ',Object.assign({},{$1') gave unexpected results when there were multiple elements with spreads (because e.g. "aba".replace(/a.*?$/, "x") returns "x") as well as attribute values containing the (sub)string ,({. This can be fixed by separately keeping track of the generated property argument.
• JSX seems to allow NUL (\u0000) in attribute values and text. As the statics were joined with NULs the parser couldn't tell apart original NULs and template variable placeholders. The joined string was also used as a cache key, with similar ambiguity problems as described in PR Use unique keys for caching results #22. This can be fixed by ensuring unique cache keys (similar to Use unique keys for caching results #22) and not joining the statics with NULs.

[jviide]

With the proposed changes the gzipped size stays just under 700 bytes (692 bytes).

[developit]

awsome, thanks for acting so quick on this! I am going to try to golf a bit post-merge but right now this is only like +2b 💃 (update: ~80b)

[developit]

@jviide hmm - I think I ran into a bug with the revised attribute logic. This test is failing:

html`<div ...${props} ...${other} />`