fix: page the ACS snapshot for holdings reads

[gabitoesmiapodo]

Summary

No related issue.

Every balance read replayed the whole ledger update stream, thousands of requests each time. That tripped the validator's rate limit, which answered with an nginx 503 page the wallet then showed to the user.

Changes

• Holdings reads page the participant's active contract snapshot instead of replaying ledger updates
• A holding whose interface view is missing now fails the read instead of being dropped from the total
• Listing holdings and summarizing them share one read

Deviations

• Grouping inside the holding summary no longer rebuilds an array per holding, since this is now the only balance path when Scan is unavailable

Acceptance criteria

• A balance read costs a handful of requests, not thousands
• A party holding more contracts than the participant returns in one response is still read in full
• Holding summaries, pending transfers and the rest of the RPC surface answer as before

Test plan

Automated tests

1. Run pnpm install
2. Run pnpm test
3. Expect 91 passing tests, no failures

Manual verification

1. On the devnet droplet, check out this branch and run docker compose up --build -d
2. On the validator host, run grep -c "POST /v2/updates" /var/log/nginx/access.log and note the number
3. Open the wallet and load a party's balance
4. Run the same count again on the validator host
5. Expect the balance to show, the count to be unchanged, and no 503

Breaking changes

None.

Checklist

• Self-reviewed my own diff
• Tests added or updated
• Docs updated (if applicable)
• No unrelated changes bundled in

Screenshots

None.