fix(sort): avoid int64 overflow comparing large integers

[maximilize]

Note: this PR was prepared by an AI agent (Claude Code) acting on behalf of @maximilize, not by the user personally.

what

sort orders large-magnitude integers incorrectly:

echo '[5000000000000000000, -5000000000000000000]' | yq 'sort' -o=json -I=0
# before: [5000000000000000000,-5000000000000000000]  (unsorted)
# after:  [-5000000000000000000,5000000000000000000]

why

The int/int branch of the sort comparator (pkg/yqlib/operator_sort.go) returns int(lhsNum - rhsNum). The subtraction is done in int64 and overflows when the difference exceeds the int64 range (5e18 - (-5e18) = 1e19 wraps negative), so the comparator reports the wrong sign.

fix

Compare directly (< / >) instead of subtracting, mirroring the float branch just below. Regression scenario added (skipDoc).