fix(jsbn2): prevent modInverse hangs on zero and negative inputs

[Kr0emer]

Summary

This PR fixes an infinite-loop DoS in BigInteger.modInverse (ext/jsbn2.js) for:

• zero input (a = 0)
• negative input (a < 0)

Root Cause

bnModInverse assumes a positive, invertible operand and directly clones this:

• a = 0 can make the loop non-terminating
• negative values are not normalized before the binary-GCD loop, which can prevent convergence

Fix

In bnModInverse(m):

• return early when m == 0
• normalize input with x = this.mod(m)
• return early when normalized x == 0
• run the existing algorithm with x (instead of raw this) in state updates

Tests

Added regression tests in test/qunit-do-crypto.html:

• BigInteger.modInverse returns quickly for zero input
• BigInteger.modInverse normalizes negative input

Validation

• Reproduced previous hang behavior on old code with timeout harness:
  • modInverse(0,3), modInverse(0,97), modInverse(-1,7), modInverse(-13,97) timed out
• After patch:
  • modInverse(0,3) -> 0
  • modInverse(0,97) -> 0
  • modInverse(-1,7) -> 6
  • modInverse(-13,97) -> 82
  • no timeout/hang
• Existing npm tests still pass in this environment.