chore(deps): bump omen_realitio_withdraw_bonds_abci CID (resume-after-timeout cache)

[jmoreira-valory]

Summary

Cascades the resume-after-timeout claim cache from valory-xyz/omen-protocol#3 (tag v0.1.2) into market-creator, for consistency with the market-resolver bump (valory-xyz/market-resolver#34).

```
third_party CID

• bafybeicj3pgdx7q5chbms3pxrb6pcv7lbqhkigwnmjuo54p63ub7mkqe5i

• bafybeiabqn6mcg4nw3mqbyxcuiyxgwlf7e2il3cj4awpmdi7bibqlwvklq
  ```

Plus the omen-protocol upstream pin bump (v0.1.1 → v0.1.2) so check-third-party-hashes recognises the new CID, and the four cascaded dev hashes (market_creation_manager_abci, market_maker_abci, agent, service). No dev source touched.

Why market-creator needs it too

market_maker_abci/composition.py wires the omen_realitio_withdraw_bonds_abci round into the creator's FSM (confirmed in composition.py, skill.yaml, behaviours.py, models.py). So market-creator runs the same bond-recovery flow and carries the same starvation risk that was fixed: under a 60s round_timeout, the RPC-heavy multisend build (eth_getLogs + eth_call simulation + safe-tx build, per claim) can run past the deadline and be discarded, restarting from scratch every cycle and never draining the claimable queue.

What the upstream change does

Caches per-question claim artifacts on SharedState.realitio_claim_build_cache. A round that times out before settling reuses the build next cycle instead of rebuilding. Each cache hit re-runs _simulate_claim to detect calldata that would now revert (subgraph lagging behind a settled claim), evicting stale entries — protecting MultiSend.delegatecall requireSuccess=true atomicity.

Validation

The fix was smoke-tested against a Tenderly Gnosis fork (real RealitioContract + skill code): a synthetic question created on the fork, answered, time-advanced to finalize, then the real _build_claim_txs cache loop driven through cold-build → warm-reuse (rebuild skipped) → post-settlement stale-eviction, with on-chain claimWinnings simulation correctly flipping True→False. All paths behaved as designed.

Test plan

• tomte tox -e isort,black,flake8,mypy,pylint,darglint — green
• tomte tox -e check-abciapp-specs,check-handlers,check-packages — green
• market_maker_abci composed-skill tests (57) pass; py3.11-linux env passes with coverage gate satisfied
• autonomy packages lock produced only the 4 expected dev cascades; third_party preserved

Rollout

After merge, bump PROPEL_SERVICE_HASH_ID in the relevant agent-deployments market-creator env(s) to the new service/valory/market_maker CID bafybeies32xs4gmyiscsvrwhh5a2koufjqzefa4y2f26sc3hbrm55knoxy, then redeploy.

[DIvyaNautiyal07]

LGTM. Verified the cascade is internally consistent and matches upstream:

• omen-protocol v0.1.2 dev CID for omen_realitio_withdraw_bonds_abci = bafybeiabqn6mcg4nw3mqbyxcuiyxgwlf7e2il3cj4awpmdi7bibqlwvklq (matches third_party entry).
• All 4 cascaded dev CIDs (market_creation_manager_abci, market_maker_abci, agent, service) appear identically in every file that references them (packages.json, aea-config.yaml, both skill.yamls, service.yaml).
• Old CID bafybeicj3pg… is fully purged; no stale v0.1.1 references in docs.
• pyproject upstream pin bumped to v0.1.2 as expected.

Nothing else to flag — pure hash-only cascade.