Tolk optimizer arithmetic overflow in ton() constants leads to incorrect values

[Gusarich]

The Tolk optimizer incorrectly computes large numeric constants passed to ton(), due to unchecked arithmetic overflow. Specifically, parse_nanotons_as_floating_string() uses signed 64-bit integers (int64_t) without overflow checks, resulting in silently corrupted numeric constants.

Example Input:

fun onInternalMessage(): int {
    return ton("10000000000");
}

Incorrect Output:

// automatically generated from file.tolk
PROGRAM{
  0 DECLMETHOD onInternalMessage
  onInternalMessage PROC:<{
    // 
    -8446744073709551616 PUSHINT	// incorrect due to overflow
  }>
}END>c

Cause:

• Internal numeric parsing (parse_nanotons_as_floating_string) uses int64_t without overflow checks.
• Multiplying large numbers exceeds the limits of int64_t, causing arithmetic wraparound and incorrect constant values.

Impact:
Compiled contracts contain silently corrupted numeric constants, potentially causing severe logic and financial errors when deployed.

Expected behavior:
Numeric parsing must directly use arbitrary-precision integers (RefInt256) or explicitly detect overflow during parsing to guarantee correctness.

---

LLM Fuzzing discovery (see tact-lang/tact#3123)