template compilation improvements

[ef4]

First, I'm refactoring so that we standardize all template compilation as inline template compilation, within Javascript, because that is the more fully-featured case. Specifically, a standalone HBS file can go through a trivial transformation to make it into a Javascript file:

<Hello />

import { hbs } from 'ember-cli-htmlbars';
export default hbs("<Hello />");

After which point we can share all the same standard machinery for compiling forward from there.

Second, I'm going to take advantage of template lexical scope in non-strict mode which we just successfully landed in ember-source 3.28.4. That will significantly streamline the build output when using staticComponents, etc, because we can pass the statically-resolved things directly into the template and completely sidestep the runtime resolver.

The first commit here is just testing out the basic idea, it needs further cleanup refactoring.

[ef4]

Reminder to self: template-compiler-broccoli-plugin.ts can drop its stage3 support, it will only need to handle stage1. Since stage one never needs the build time resolver, this may result in a much simpler setup that doesn't need our whole NodeTemplateCompiler at all.

[ef4]

These changes can also address a cache-safety hole we currently have: the stage1 TemplateCompilerBroccoliPlugin uses broccoli-persistent-filter but only caches off:

• hashDeps of @embroider/compat
• the checksum of ember-template-compiler.js

It doesn't incorporate a checksum over the set of plugins.

[ef4]

I'm going to land this long-running branch, it has a good refactor and all tests are passing.

The actual feature work on top to use lexical scope can be a followup.