CVE-2023-25653

### Description

‍
When using the non-default "fallback" crypto back-end, ECC operations in `node-jose` can trigger a Denial-of-Service (DoS) condition, due to a possible infinite loop in an internal calculation.  For some ECC operations, this condition is triggered randomly; for others, it can be triggered by malicious input.

#### Technical summary

The JOSE logic implemented by `node-jose` usually relies on an external cryptographic library for the underlying cryptographic primitives that JOSE operations require.  When WebCrypto or the Node `crypto` module are available, they are used.  When neither of these libraries is available, `node-jose` includes its own "fallback" implementations of some algorithms based on `node-forge`, in particular implementations of ECDH and ECDSA.

A various points, these algorithm implementations need to compute to the X coordinate of an elliptic curve point.  This is done by calling the `getX()` method of the object representing the point, which is an alias of the function `pointFpGetX()` in `lib/deps/ecc/math.js`.

Computing the X coordinate from the form in which the point is stored requires computing the modular inverse of the Z coordinate, using the `modInverse` function from the `jsbn` library (e.g., `this.z.modInverse(this.curve.p)`).  The output of this function call is multiplied by another value before being reduced with the `barrettReduce()` function.

The root cause of this issue is that the `jsbn` `modInverse` function sometimes returns negative results.  These results are correct in that they are equivalent mod the relevant modulus, but can be problematic for functions that expect modular operations to always return positive results (in the range `[0, p)`, where `p` is the modulus).

In particular, while the Barrett reduction algorithm in general can handle negative inputs, the implementation in `node-jose` explicitly does not. Therefore, while the negative value that is returned by `modInverse()` is mathematically correct, it leads to an error in `barrettReduce()` causing an infinite loop which may result in a Denial of Service condition.

For a given prime modulus, we estimate that roughly one in every `2^20` inputs produce a negative `modInverse()`.  This estimate was validated with exhaustive testing on small primes (