Remove reverse readline, test against NP1 and recover NumPy 1 dependency support

[DanielYang59]

Summary

• Replace monty reverse_readline with built in readline
• release numpy >2 pin
• test numpy 1 https://github.com/CederGroupHub/chgnet/pull/203/checks?check_run_id=30164697551

I'm not seeing much point in using reverse_readline as all lines are read, and reversed immediately:

chgnet/chgnet/utils/vasp_utils.py

Lines 61 to 67 in 11ff513

	charge, mag_x, mag_y, mag_z, header, all_lines = [], [], [], [], [], []
	for line in reverse_readfile(outcar_path):
	clean = line.strip()
	all_lines.append(clean)
	all_lines.reverse()

[DanielYang59]

@janosh I believe this one is ready for review. I completely removed usage of reverse_readline because that OUTCAR file is read in entirely and reversed right after read in, as such I'm not seeing much point to use reverse_readline.

The reverse readers from monty is much slower than the built in readline(understandably because that function from monty is 11 years old and was never updated), see the benchmark in materialsvirtuallab/monty#712 (comment). I would expect reverse readers to be more RAM-efficient though (but still bring no benefit to the use case here).

[janosh]

good thing you noticed. strange that it read the file in reverse only to reverse it again right away. seems like unnecessary work. was that leftover by accident @BowenD-UCB?

[janosh]

@DanielYang59 given we're no longer reliant on materialsvirtuallab/monty#712, chgnet should be ready for a new pypi release with numpy 2 support. anything i'm missing?

[DanielYang59]

chgnet should be ready for a new pypi release with numpy 2 support.

I believe we are :) Nothing missing AFAIK