fix script argument regression, add process.argv.lsc

[rhendric]

In commit 03c6e6a, removing the code that mutated process.argv also removed a side effect that prevented all arguments to lsc from being handled as scripts to compile and run. This caused a regression when running lsc script-file.ls arg1 arg2 that would result in a crash when no arg1.ls file could be found.

The regression is fixed by reintroducing the side effect to remove all but the first positional argument from the array containing the scripts to compile and run, but only when lsc is running scripts—batch compilation commands involving multiple input files should remain unaffected.

This commit also adds a process.argv.lsc property, present only when a script is being interpreted by lsc, which is an array holding the name of the script being invoked along with any extra arguments not consumed by lsc. The intent here is to provide a way to get script arguments that's simpler and more reliable than picking through process.argv and skipping over elements like node, the lsc compiler itself, and any CLI options being provided to either.

---

Fixes #1013. This is both a bug fix and an experimental feature (the process.argv.lsc part), so I would like to hear some feedback on whether this mild alternative to the original process.argv mutation that I nuked back in #959 is reasonable—if not, it wouldn't be hard to remove that part and just make this the bug fix.

[rhendric]

Crickets and thumbs up, so I'm issuing final call for comments on this. Will merge on or after Friday, March 23 if nobody has anything to say.

[corwin-of-amber]

Passing process.argv.lsc to command line argument parsers (e.g. commander) results in an error since these in fact expect the node executable to be the first argument. So I have to do something like:

opts.parse (if process.argv.lsc then ['', ...process.argv.lsc] else process.argv)

[rhendric]

@corwin-of-amber, that's an interesting point. Part of why I wanted to add the .lsc property instead of continuing to mutate process.argv itself was that I didn't think there was a good way to uniformly adhere to the various expectations that different libraries placed on process.argv; this feels like another one of those. But maybe there's little harm in adding a dummy entry to the start of process.argv.lsc just so that it can easily be used as a substitute in this particular use case. It strikes me as a hack, but maybe it's worth it?

[corwin-of-amber]

Sounds to me like process.argv.lsc[0] should be lsc itself – as it is the interpreter running the current script (similar to bash, python, node etc.).