Humdrum Typing and Improvements

[mscuthbert]

Wanting to solve #62 (and #1882) means understanding how Humdrum Spine Parsing works, a topic that I last looked at around 2010. So before changing the code to something new, need to get it back up to date. AI-assisted PR to clean up the code:

1. Type annotations throughout.
2. Removed legacy patterns — like weakrefs for spineCollection and manual __iter__/__next__/iterIndex, is True/is False checks, _elements private access, class.__class__ swaps without running __init__.
3. Stop using "position" -- a meaningless term. Use lineNumber (yes, it's 1-based sadly, but that's how it goes). fileLength replaced by numLines property.
4. Real bar.Repeat objects instead of poking a non-existent Barline.repeatDots.
5. WeakKeyDictionary humdrumLineNumbers maps M21Objects to their line numbers rather than adding and removing attributes.
6. Renamed moveDynamicsAndLyricsToStreams → attachNonKernEvents and consolidated a lot of branches (that routine already did much more than just Dynamics and Lyrics).
7. Tuplet sums use opFrac; redundant integer-ratio fallback removed.
8. Removed flag attributes isSpineLine, isComment, isGlobal, isReference from HumdrumLine and subclasses; use isinstance() instead and get type narrowing.
9. Add _reprInternal to most classes (and make remaining also be ProtoM21Objects)
10. Blank lines (technically illegal in Humdrum, but seen often in practice) are stored so that line numbers between files and eventLists remain constant (-1)

[coveralls]

coverage: 93.07% (+0.02%) from 93.052% — humdrum-typing into master