Optimize for the common case (#17)

* Generate a memoized function for computing a style's name

* Use the newly-generated naming fn; improve manual :key support

* Clean up

* Prepare to only conditionally compile CSS, update DOM

* Respect global :always-compile-css?

* Add support for :always-compile-css per-function escape hatch

* Clean up some TODOs

* Change StyleContainer to support skipping compile when mounted

Now for the common case of applying a style that's already in the DOM,
all we have to do is compute its name (which is memoized on the params!)
and lookup the info to handle any composed styles. We don't run the
style factory fn and we certainly don't compile any CSS!

* Invoke style-factory directly, without `apply`

* Update core-test to remove unnecessary apply params

* Fix: find-key-meta not working consistently

* Rename vars for consistency

Use a separate repository name for CI snapshots

Fix support for passing a list of composed styles

Update release script to python3

Fix missing style names when compiled under :simple optimizations

Fixes #7

Don't make style names that start with -

Can happen under advanced compilation if the fn's name starts with _

Update package-lock.json

Don't generate cljdoc for the util namespace

Probably could hide the runtime namespace, too...

Add parameters to defkeyframes (#5)

* Make defkeyframes accept a params vector

This is a breaking change! But it makes the API much more logically
consistent—why shouldn't you be able to parameterize keyframes?

This change also includes some heavy refactoring to simplify code
generation for the other cases, as well as an optimization that skips
the call to build-style-name if the function does not accept parameters.

* Update README to reflect params support on defkeyframes

Add more print output to the version