feat: derive EIP-8198 slot duration from fork config

[barnabasbusa]

Updates dora to consensus-specs v1.7.0-beta.4. #5710 replaces the standalone SLOT_DURATION_SCHEDULE with a fork-specific SLOT_DURATION_MS_EIP8198, so dora now derives the wallclock schedule from SLOT_DURATION_MS at genesis plus SLOT_DURATION_MS_EIP8198 from EIP8198_FORK_EPOCH. go-eth2-client is bumped to ethpandaops/go-eth2-client#64 for the Heze EIP-8015 containers (no eth1 data / deposits), and the state transition follows the beta.4 changes: pending builder payments of any slashed proposer are cleared in slash_validator (#5719), the builder sweep withdraws the balance left after earlier withdrawals in the same payload, and Heze skips deposits that would create validators with BLS withdrawal credentials (EIP-8365). The beta.4 vector replay passes again and now covers Heze.

🤖 Generated with Claude Code

[redpandabot]

The PR follows consensus-specs v1.7.0-beta.4: the EIP-8198 slot schedule is now derived from the scalar SLOT_DURATION_MS (genesis) and SLOT_DURATION_MS_EIP8198 (from EIP8198_FORK_EPOCH), with the wallclock seeding a genesis entry and the old SLOT_DURATION_SCHEDULE handling removed. I traced the derivation, duplicate/overflow handling in newWallclockSchedule, and the CheckMismatch slice branches, and found the change consistent and correct. No real problems to report.

---

_{Reviewed 3 changed file(s) @ 79e1bf07 — no blocking issues found.
A red panda's scientific name, Ailurus fulgens, means "shining cat."}

[pk910]

SlotDurationScheduleEntry should be removed entirely.
It's a leftover as AI doesn't like removing code. Tell it to clean up stuff ;)
Pass the specs to the relevant codepaths that need the schedule, not a separate parsed schedule array

[barnabasbusa]

Removed SlotDurationScheduleEntry and GetSlotDurationSchedule entirely in c9ea455. SetupClock / newWallclockSchedule now take the ChainSpec and build the segments straight from SLOT_DURATION_MS + EIP8198_FORK_EPOCH/SLOT_DURATION_MS_EIP8198; the generic sort/merge logic is gone and the tests use specs too.

[barnabasbusa]

Also dropped wallclockSegment in c83e4e8: the schedule is now flat (slotDuration, forkSlot, forkSlotDuration and a single precomputed forkOffset), and all the segment loops are unrolled into a before/after-fork branch.

[barnabasbusa]

Reverted the segment flattening again (force-pushed back to b896613): with wallclockSegment it reads cleaner, so the schedule stays specs-driven but keeps the segments.