Loss support (better version: Pauli envelop)

An upgrade from naive random imputation (#75 ): https://arxiv.org/pdf/2603.04156

Is your feature request related to a problem? Please describe.
Qubit loss is a critical feature for atom/ion platforms, but currently deq can only handle Pauli noises. Although in principle deq provides the flexibility of changing probabilities at runtime, there is no such tool that can easily do it. Ideally, user would want to declare loss noise in the gadgets (as an extension to stim), and then deq should absorbs all the complexity inside, instead of requiring user to provide probability modifiers.

Describe the solution you'd like
Adopt the state of the art loss decoding technique: Pauli envelop. Need more in-depth analysis under the context of dynamic logical circuits for a systematic implementation.

Describe alternatives you've considered
Alternative would be to treat loss as erasure and simply set weight to 0 of some edges. This would be suboptimal but should be an alternative approach that's available for benchmarking.

Start by reading the linked Pauli-envelop paper and issue #75, then analyze how loss should work with dynamic logical circuits. Compare the proposed Pauli-envelop approach with the erasure alternative and define what a systematic implementation and benchmarking path would require.