JJava 1.0-a7 released

[andrus]

Glad to announce a new a7 milestone of DFLib JJava 1.0. JJava is a Java / JVM kernel for Jupyter. The new release focused on a more robust kernel boot sequence and, more generally, code execution flow, as well as the %time magic. Key improvements:

1. Explicit evaluation on startup of the script coming from JJAVA_STARTUP_SCRIPT. Previously, it was executed lazily within the notebook that resulted in random issues like `JJAVA_STARTUP_SCRIPT can't start up spark #67.
2. The ability to declare mutually-dependent classes. Something that the underlying JShell could do properly, but JJava would choke on due to the incorrect handling of recoverable conditions. This should work now.
3. Reworked the kernel code execution flow for clarity. The new API uses kernel.evalBuilder(code) method to start a configurable execution.
4. Reworked the execution thread model. If backend JShell code invocation happens from another JShell invocation, it will be run in the current thread instead of a separate thread. This (a) makes sense in general and (b) makes invocation telemetry much easier to implement (see below)

Ok, kernel internals may be fun for some of us, but a general user couldn't care less. So here is another new feature that will hopefully get more people excited ...

5. %time / %%time magic. This is a handy utility to take quick measurements of the execution time of a given piece of code in the notebook. In most cases I prefer the "cell" flavor (%%time), using it like this:

%%time
for(int i = 0; i < 5; i++) {
    Parquet.loader().load("large-file.parquet");
}

CPU times: user 3.204 s, sys 0.068 s, total 3.272 s
Wall time: 3.289 s

[andrus]

@maxandersen , note that there were some kernel API changes in the latest JJava release. This made the internals a bit more sane. But I hope they can be ported easily to your JBang subclass (specifically, the item 3 above kernel.evalBuilder(..)).