Header read-only

[Ghesselink]

only_header limits what is returned, not what is read: currentlyparse() reads the whole file and the comment-stripping regex makes a second full copy, before the header (a few hundred bytes, always at the top per ISO 10303-21) is sliced out.

This PR stops reading after the ENDSEC; that closes the HEADER section (64 KiB chunks).

Some tests: peak RSS of an only_header=True parse:

File size	Before	After
12 KB	22.6 MB	22.6 MB
70 MB	157.9 MB	22.6 MB
130 MB	275.2 MB	22.6 MB

Also slightly faster on large files, but it was never a time cuonsumer to begin with (0.37 s → 0.19 s at 130 MB).

Testing: test_parser.py tested with VS testfiles from the header validation, outcome is the same. g

[aothms]

I don't spot anything wrong with it, but I also find it a bit of a weird compromise.

I would say:

• Either, build our own streaming lexer and terminate as soon as we have a complete header. I think this is overkill
• Just take the first 1 megabyte (or 10 megabyte if you are crazily careful) unconditionally and operate on that.