BUG: Fix segfault in read_csv with extremely large exponents

[ssam18]

Closes #63089

Description

This PR fixes a segmentation fault that occurs when reading CSV files containing numbers with extremely large exponents in scientific notation (e.g., 4e492493924924).

Root Cause

The issue was an integer overflow in the xstrtod() function in pandas/_libs/src/parser/tokenizer.c. When parsing the exponent portion of scientific notation, the code accumulated digits into an int variable without bounds checking:

int n = 0; while (isdigit_ascii(*p)) { n = n * 10 + (*p - '0'); // Integer overflow with large exponents num_digits++; p++; }

With an exponent like 492493924924, the variable n would overflow, causing undefined behavior that manifests as a segmentation fault.

Solution

I added a maximum digit cap (MAX_EXPONENT_DIGITS = 4) when accumulating the exponent value:

• Only the first 4 digits are used for the actual exponent value (allowing up to 9999)
• Remaining digits are still consumed to maintain correct parsing position
• This is sufficient since valid double-precision exponents are limited to roughly ±308 anyway
• The existing range check (DBL_MIN_EXP to DBL_MAX_EXP) will properly handle out-of-range values

Testing

Added test_issue_63089.py with test cases covering:

• The exact case from the issue report
• Various edge cases with extremely large positive and negative exponents
• Numbers with decimal points and large exponents

The fix prevents the overflow while maintaining correct parsing behavior for valid scientific notation.

---

Checklist:

• Closes BUG: read_csv results in SIGSEGV #63089
• Added test case
• All existing tests should pass (haven't run full suite locally due to build dependencies)