fix(sequencer): avoid sticky min fee L1 read failures

[AztecBot]

Summary

• compute the next-slot timestamp locally in FeeProviderImpl instead of making an extra rollup getTimestampForSlot read
• reset the cached current-min-fee refresh after rejected L1 reads so later retries can recompute instead of replaying the same rejected promise
• add regression coverage for local timestamp calculation and retry after transient L1 failures

CI failure investigated

• http://ci.aztec-labs.com/1782904915615458
• failing sublog: http://ci.aztec-labs.com/66ef70e6044f13bd (docs/examples/bootstrap.sh execute)
• observed failure: repeated wallet getMinFees calls surfaced node-side rollup reads failing for getTimestampForSlot(54) / getManaMinFeeAt, causing example_swap and recursive_verification to fail after retries

Verification

• yarn workspace @aztec/sequencer-client test src/global_variable_builder/fee_provider.test.ts
• yarn format --check sequencer-client
• yarn lint sequencer-client
• git diff --check

Local CI limitation

• ./bootstrap.sh ci is not a valid root target in this checkout (Unknown command: ci)
• ./bootstrap.sh build yarn-project could not complete here because this session cannot use Docker and Noir bootstrap could not fetch just@1.42.4 from crates.io; yarn build therefore fails earlier on missing generated protocol/L1 artifacts before reaching this patch

---

Created by claudebox · group: slackbot