chore: bump mech to v0.35.2

[DIvyaNautiyal07]

Summary

Bumps valory-xyz/mech upstream pin from v0.35.1 to v0.35.2.

Release: https://github.com/valory-xyz/mech/releases/tag/v0.35.2

Includes:

• mech#486 — fix(task_execution): coerce request_ids to bytes32 for batch status check. Fixes the silent 500 on fetch_batch_request_id_status under eth_abi>=6, unblocks marketplace pending-queue reconciliation, and stops the mech-watcher-driven force-restart loop on affected marketplace mechs.

Test plan

• autonomy packages lock --check OK
• CI

🤖 Generated with Claude Code

[jmoreira-valory]

Review round 1 — ✅ Approve

What this does. Bumps the upstream valory-xyz/mech pin from v0.35.1 to v0.35.2 in mech-predict. That means new CIDs for the three vendored mech skills (mech_abci, task_execution, task_submission_abci), cascaded through the agent's aea-config.yaml and on into the mech_predict agent and service hashes, plus the upstream_pins entry in pyproject.toml. Functionally it picks up mech#486, which stops fetch_batch_request_id_status from silently 500-ing on the marketplace pending-queue reconciliation (the failure that was driving the mech-watcher force-restart loop).

Verified against source at HEAD rather than trusting the PR body:

CID correctness — diffed mech@v0.35.1...v0.35.2's own packages/packages.json. Exactly five entries moved upstream: mech's own agent/service (not consumed here) and the three skills. All three new CIDs in this PR match the v0.35.2 values byte-for-byte.

No partial bump — compared all 41 third_party entries in packages/packages.json against mech@v0.35.2: 41/41 identical, zero drift. Nothing that should have moved was left behind.

Cascade complete — the repo has exactly two package configs (agents/mech_predict/aea-config.yaml, services/mech_predict/service.yaml), and the bottom-up chain is intact at every rung: skill CIDs -> aea-config.yaml -> agent/valory/mech_predict -> service.yaml agent: pin -> service/valory/mech_predict. The 28 custom/... tool CIDs correctly stay put — they sit outside the agent dependency graph. No stale v0.35.1 CID survives anywhere in the tree, and pyproject.toml:122 upstream_pins is the repo's only mech version pin (tox.ini, Makefile, Dockerfile.verify, uv.lock and the workflows carry none).

No config-surface change to absorb — the v0.35.2 delta is only mech#486. The three skill.yaml diffs upstream are fingerprint/dependency-CID lines only: no new or renamed models.params.args key, so there is no service.yaml override-parity work and no AEAEnforceError startup risk. mech_abci's FSM is untouched (no rounds.py / fsm_specification.yaml in the compare), so no 4-place sync drift. Release v0.35.2 is published, non-draft, non-prerelease, and its tag is identical to mech's main.

CI — 23 green, publish skipped, integration_tests still in_progress at draft time (that job is green on recent main commits). This approval assumes it lands green; if it fails, it should be looked at before merge.

Follow-up, not a blocker — merging this does not put the mech#486 fix on the mechs. It needs a mech-predict release, then the PROPEL_SERVICE_HASH_ID bump in agent-deployments for the affected marketplace mechs, since the service/valory/mech_predict CID changed here.

No findings.