[tetsuo-cpp]

The parser converted number tokens with std::stoll and std::stod. Both throw std::out_of_range when the value does not fit, and nothing caught it, so the compiler aborted.

Change

• Integer literals are parsed with StringRef::getAsInteger (a leading + is stripped first, as isNumber allows it). Out-of-range values report integer literal out of range for i64 at the token.
• Float literals are parsed with llvm::to_float, the same routine the lexer uses to classify them. A value that overflows to infinity reports float literal out of range for f64. Underflow is accepted and rounds to zero or a subnormal, as in C. Before this change underflow also crashed, because std::stod throws on any ERANGE.
• The lexer's classification logic is unchanged (Hex literals, nan and inf are silently lexed as float literals #112 touches isFloat separately).

Before

$ warpforthc int.forth   # 99999999999999999999 a !
libc++abi: terminating due to uncaught exception of type std::out_of_range: stoll: out of range
exit=134
$ warpforthc flt.forth   # 1e999 a !
libc++abi: terminating due to uncaught exception of type std::out_of_range: stod: out of range
exit=134

After

$ warpforthc int.forth
int.forth:3:1: error: integer literal out of range for i64
99999999999999999999 a !
^
error: failed to parse Forth source
exit=1
$ warpforthc flt.forth
flt.forth:3:1: error: float literal out of range for f64
1e999 a !
^
error: failed to parse Forth source
exit=1

Tests

• New: integer-literal-overflow-error.forth, integer-literal-underflow-error.forth, float-literal-overflow-error.forth. Each one crashed before this change.
• basic-literals.forth now covers +5, INT64_MAX and INT64_MIN.
• float-literals.forth now covers DBL_MAX and 1e-400, which underflows to zero.

Fixes #111