Stop the sum filter rounding integer totals #18

The sum filter uses floating-point accumulation for integer inputs, which can silently change the total.

{{ [9007199254740993, 2] | sum }}

Python Jinja2 returns:

9007199254740995

The affected Go implementation returns:

9007199254740994

Every input and the correct total fit within the supported integer range, so this isn’t an overflow case. We’re losing precision by taking integer arithmetic through floating point.

I’m preparing a PR to use the engine’s existing value-addition operation for accumulation. That keeps integer inputs in integer arithmetic and allows floating-point promotion when required. I’ll attach it here with a regression test checking the exact rendered result. This change won’t add arbitrary-precision integers or address integer overflow beyond the engine’s existing behaviour.