Store Cashu proofs in one blob

[benthecarman]

Persist the complete Cashu proof set with one versioned KV write so VSS-backed wallets avoid a round trip per proof. This should make startup much faster for VSS based cashu wallets.

[hash-money]

Tested at head 0499e76 against our staging VSS (vss-server behind Caddy, Mutinynet signet, self-hosted Cashu mint) — same setups as the numbers in #90. Two environments: desktop wallet-cli (wired; ~0.6 s cold TLS+HTTP round-trip to the VSS host) and the Samsung A12 on a mobile network. Desktop wallets used fresh seeds; the A12 ran the PR build over its existing long-lived wallet. Workload: 5,000-sat trusted receive (37 proofs), paid by a third node.

Desktop (wallet-cli, fresh wallet, one 5k receive in store):

Metric	4e9d230 (base)	PR #93
Cold restart → API ready	78.5 s	18.2 s
Receive: paid → PaymentReceived	~64 s	~30 s
get-invoice (mint quote)	10.7 s	11.5 s (unchanged)

(Local-store control on the PR build: get-invoice 1.2 s, receive ~1 s.)

A12 (existing wallet with channels; baseline re-measured the same afternoon, minutes before the upgrade):

Metric	base	PR #93
Cold start, pre-LDK phase	163.6 s (matches the 164 s in #90)	31.6 s (5.2×)
Receive: paid → PaymentReceived	~65 s	37.3 s
Receive-URI generation	11.95 s	11.8 s (unchanged)

Reading of the results: startup improves ~4–5× (proofs dominated the key count, and the N serialized reads collapse to one), the receive saga roughly halves (proof writes collapse to one blob write), and the two paths that never touch proofs are — as this PR would predict — unchanged: quote issuance still serializes the mint-quote/counter/saga-state writes, so the remaining gap to local-store (~1 s) is those per-key round-trips plus the bitreq connection churn (rust-bitcoin/corepc#658 / #659).

One thing to flag before this ships: there's no migration from the legacy per-proof keys. load_proofs_from_store reads only proof_set and treats NotFound as an empty proof set, so an existing wallet upgrades into a silently-empty trusted balance:

• desktop: a base-build wallet holding 5,000 sats read 0 sats under the PR build; reverting the binary showed 5,000 again (legacy keys untouched — recoverable, but silent);
• A12: the existing wallet's trusted balance vanished from the UI after upgrade, and the subsequent receive minted into the blob — the store is now forked (new build can't see legacy proofs, old build can't see the blob).

A one-time fallback — blob key missing → list + read legacy proof_* keys → write the blob → delete the legacy keys — would close that, and as a bonus deletes the very keys that make today's startup slow. Happy to re-run this capture on any update.

[benthecarman]

Do you actually need the migration? I excluded it because i assumed no one is using this on mainnet yet

[hash-money]

No — we don't need it. Our Cashu tier is signet-only staging and the affected wallets are disposable test wallets; we'll recreate them when we bump our pin to a rev that includes this. From our side this is good to go as-is — the numbers above stand.