Refactor LazySegmentTree to be easier to modify

[Mukundan314]

Haven't tested speed or correctness yet

[Mukundan314]

Example usage: https://codeforces.com/contest/2027/submission/290258327 (lazy segment tree is overkill for this problem but demonstrates the usage)

[cheran-senthil]

Updated this branch onto current master and finished the refactor around the source-specialization model demonstrated in the linked solution.

What changed:

• preserved the existing constructor, range-add/range-min/max behavior, and verifier API
• kept the four operation-specific hooks (_unset_lazy, _apply_to_data, _apply_to_lazy, _merge_data) and the buffer slot, so multi-field state can be implemented by editing those small sections
• exposed apply() while retaining add() as an alias
• added randomized tests for the default min/max behavior and a minimum-with-count specialization matching the D2 use case

Correctness:

• full suite: 63 passed
• the supplied D2 solver was separately checked against brute force on 48,000 random small instances

Performance on 50,000 mixed operations with n = 32768 (best of 7):

Runtime	master	this branch	delta
CPython 3.13	0.604s	0.808s	+34%
PyPy 3.11 / 7.3.23	0.0365s	0.0383s	+5%

I also benchmarked runtime callbacks and a fully generic action/composition interface; those were roughly 2x or worse for the actual minimum-with-count specialization, so I left them out. Local binding of the hook methods also made PyPy slower, so the hot loops retain direct method dispatch.