Allow wallet policies with `musig()` to pre-generate nonces before the transaction is known

[bigspider]

Currently, the psbt_session_id (and the entire MuSig2 Round 1 flow) are dependent on the actual transaction being signed.

This prevents software wallets that want to benefit from the non-interactive signing with preprocessing, which would allow the UX of MuSig2-based wallet policies to practically be a single round like a normal multisig wallet policy (by pre-executing Round 1 when given the chance - for example right after sending a transaction).

The coordinator software wallet would therefore need to pregenerate the pubnonces (after deciding a maximum number of inputs - as large as possible, but taking into account what is feasible on embedded devices), and then directly execute Round 2 once the transaction is known.

For that to happen, we would need some changes in the app:

• change the psbt_session_id to only depend on the wallet policy, and not on the transaction
• modify musig_nonce_gen - currently it depends on the aggregate key tweaked for the input pair (base don (is_change, address_index); it could rather depend on the untweaked aggregate key.
• make it easy and cheap to execute Round 1. Currently, even in Round 1, the app does the full inputs/outputs validation to check whether inputs are internal, which is expensive. All those checks should be skipped for pubnonce-only generation. We can probably figure out how to signal 'MuSig Round 1 only' from the PSBT in order to avoid changes in the wire protocol, but we need to figure out the cleanest way.

To consider:

• Making the psbt_session_id only depend on the wallet policy would have the side effect of only allowing a single signing request to happen for that account. It might be worth allowing the software wallet to optionally provide an ID and committing to it in psbt_session_id, allowing multiple sessions per account managed by the same (or multiple) coordinators.