Well, if we assume that:
- $e$ is prime (65537 is)
- Only one of the primes minus one has $e$ as a factor; for example, $p-1$ is divisible by $e$, but $q-1$ is not. For this discussion, we'll assume that $p$ is the prime with $p-1 \equiv 0 \bmod e$ (which might happen to be the size 1023 factor for you)
- $p-1$ is not divisible by $e^2$
- That the ciphertext was actually generated by computing $P^e \bmod n$ for some plaintext value $P$.
Then, one way to derive the possible plaintexts is to compute:
$$C^d \cdot L^i \bmod n$$
where:
- $C$ is the ciphertext
- $d = e^{-1} \bmod \lambda / e$ . This is well defined, as $\lambda/e$ is an integer which is relatively prime to $e$.
- $L = k^{\lambda/e} \bmod n$, where $k$ is an integer such that $L \ne 1$ (and any such value $L$ works); most values of $k$ work
- $\lambda = (p-1)(q-1)/\gcd(p-1, q-1)$
- $i$ is any integer $0 \le i < e$
Now, if we iterate over the possible values of $i$, this will give $e$ possible values for the plaintext (unless $C$ happens to be a multiple of $p$). The original plaintext will be one of these values. All these values, when raised to the power $e$, will result in the ciphertext, hence we cannot distinguish from the ciphertext which one it is.