Cached dataframe diff never detects unchanged columns, and lists_are_equal only checks the last element

[nikhilsi]

Two related comparison bugs in the caching stack at commit 7952236. Filing them
together since both live in the cache diff/signature machinery. (The malformed
clear_cache path templates in the same file already have a fix in open PR #364;
this issue covers the two remaining problems.)

1. In write_cached_df (openavmkit/utilities/cache.py, lines 296-310), the
   column-modified test is unsatisfiable:

values_equal = col_new.values == col_orig.values
na_equal = col_new.isna() & col_orig.isna()
count_na_equal = na_equal.sum()
count_values_equal = values_equal.sum()
count_to_match = len(col_new)
all_equal = (
    count_na_equal == count_to_match
    and count_values_equal == count_to_match
)

A NaN position can never count as values-equal and a non-NaN position can never
count as na-equal, so the two counts cannot both reach len(col_new) for any
nonempty column. Every common column is therefore flagged as modified and
rewritten on every save, defeating the incremental diff. Fix: a position
matches if it is values-equal OR both-NaN:

all_equal = bool((values_equal | na_equal.to_numpy()).sum() == len(col_new))

2. lists_are_equal (openavmkit/utilities/assertions.py, lines 104-107)
   overwrites result on every iteration, so only the last element decides:

for i in range(len(a)):
    entry_a = a[i]
    entry_b = b[i]
    result = objects_are_equal(entry_a, entry_b)

lists_are_equal([2, 2], [3, 2]) returns True. This helper backs
dicts_are_equal, which the cache uses for signature matching (cache.py line
510), and several test assertions. Fix: break with False on the first mismatch.