Yet another refactoring to untangle the old mess.
Highlights:
base.ts, pyscript.ts and pyrepl.ts were a tangled mess of code, in which each of them interacted with the others in non-obvious ways. Now PyScript is no longer a subclass of BaseEvalElement and it is much simpler. I removed code for handling the attributes std-out and std-err because they are no longer needed with the new display() logic.
The logic for executing python code is now in pyexec.ts: so we are decoupling the process of "finding" the python code (handled by the py-script web component) and the logic to actually execute it. This has many advantages, including the fact that it will be more easily usable by other components (e.g. pyrepl). Also, note that it's called pyexec and not pyeval: in the vast majority of cases in Python you have statements to execute, and almost never expressions to evaluate.
I killed the last remaining global store, scriptQueue tada. As a bonus effect, now we automatically do the correct thing when a <py-script> tag is dynamically added to the DOM (I added a test for it). I did not remove svelte from packages.json, because I don't fully understand the implications: there are various options which mention svelte in rollup.js and tsconfig.json, so it's probably better to kill it in its own PR.
pyexec.ts is also responsible of handling the default target for display() and correct handling/visualization of exceptions. I fixed/improved/added display/output tests in the process.
I also found a problem though, see issue #878, so I improved the test and marked it as xfail.
I removed BaseEvalElement as the superclass of most components. Now the only class which inherits from it is PyRepl. In a follow-up PR, I plan to merge them into a single class and do more cleanup.
During the refactoring, I killed guidGenerator: now instead of generating random py-* IDs which are very hard to read for humans, we generated py-internal-X IDs, where X is 0, 1, 2, 3, etc. This makes writing tests and debugging much easier.
I improved a lot our test machinery: it turns out that PR #829 broke the ability to use/view sourcemaps inside the playwright browser (at least on my machine).
For some reason chromium is unable to find sourcemaps if you use playwrights internal routing. So I reintroduced the http_server fixture which was removed by that PR, and added a pytest option --no-fake-server to use it instead, useful for debugging. By default we are still using the fakeserver though (which is faster and parallelizable).
Similarly, I added --dev which implies --headed and also automatically open chrome dev tools.
As the title stays, the main goal of the branch is to kill the infamous runtimeLoaded global store and all the complications, problems and bugs caused by the fact that in many places we needed to ensure/wait that the global runtime was properly set before being able to execute code.
The core idea is that runtime is never a global object and that it's passed around explicitly, which means that when a function receives it, it is guaranteed to be initialized&ready.
This caused a bit of complications in pybutton.ts, pyinputbox.ts and pyrepl.ts, because they indirectly want to call runtime.run from connectedCallback, which is the only place where we cannot explicitly pass the runtime because it's automatically called by the browser.
But also, it is also a sign of a bad design, because it were entirely possible that connectedCallback was called before the runtime was ready, which probably caused many bugs, see e.g. #673 and #747.
The solution to is use dependency injection and create the class later on: so instead of having a global PyButton class which relies on a global runtime (whose state is uncertain) we have a make_PyButton function which takes a runtime and make a PyButton class which is tied to that specific runtime (whose state is certainly ready, because we call make_PyButton only when we know that the runtime is ready).
Similar for PyInputBox and PyRepl.
Other highlights: thanks to this, I could kill the also infamous runAfterRuntimeInitialized and a couple of smelly lines which used setTimeout to "wait" for the runtime.
While I was at it, I also called a lot of other stores which were completely unused and where probably leftovers from a past universe.
* Add unit tests for pyloader and pytitle
* Add more unit tests for pyrepl, pybox and pyinputbox
* Add more tests for pyscript and pyconfig
* White space
* Fix d3 tests and improve more examples test
* Update matplotlib test
* Add numpy to dependencies
* Address Madhur comments
* Update test name
