fix(core): promote to BigInt instead of panicking on neg/abs of i64::MIN 💥

[timfennis]

Context

Continuing the recent sweep of panic-on-edge-case fixes (negative power, windows size 0, integer division by zero), I went looking for another arithmetic input that crashes the interpreter. Negating or taking the absolute value of i64::MIN does it: the positive magnitude 2^63 does not fit in an i64.

abs(-9223372036854775807 - 1)   // i64::MIN

Int::abs and Int::neg called i.abs() / i.neg() directly on the Int64 variant. Both overflow at i64::MIN, panicking with "attempt to negate with overflow" in debug builds and silently wrapping to a negative result in release.

Changes

• ndc_core/src/int.rs: Int::abs and Int::neg now use checked_abs / checked_neg and promote to BigInt on overflow, matching the checked-then-promote pattern already used by the binary operators.
• Regression tests: 001_math/032_neg_abs_i64_min.ndc and 900_bugs/bug0026_neg_abs_i64_min_overflow.ndc.

Every other path produces the same results as before; only the i64::MIN boundary changes from a panic to the correct 9223372036854775808.

🤖