fix: compare poll endBlock against L1 block number instead of L2 on Arbitrum

[adamsoffer]

Summary

• Poll tally handlers (updatePollTallyOnReward, OnBond, OnUnbond, OnRebond) compare poll.endBlock against event.block.number to check if a poll is still active. On Arbitrum, poll.endBlock is on the L1 block scale (~24M) since Solidity's block.number returns L1-approximate values, but event.block.number in The Graph is the L2 sequential block (~437M). The check always evaluates as "poll ended," so voteStake never updates after the initial vote.
• Replace event.block.number with getBlockNum() (which calls roundsManager.blockNum() and returns the L1 block number) in all three endBlock checks so both sides of the comparison are on the same scale.

Evidence

Comparing voteStake (Vote entity) vs totalStake (Transcoder entity) for orchestrators with nonVoteStake: 0 on the current active poll (0x9bddf5c1766924e5ea601eb0bb00dc022f5980d0):

Orchestrator	voteStake	totalStake	Gap	Voted
0xbd677e...22bd	368,095	369,482	+1,387	02/26
0x5d98...5e60	413,586	415,239	+1,652	02/27
0x9d61...e3c8	287,696	287,696	0	03/02

The gap grows proportionally with time since the vote (accumulated daily rewards). Reward events fire and update Transcoder.totalStake correctly, but the poll tally handlers skip the voteStake update because they think the poll has already ended.

Test plan

• Verify subgraph builds (yarn prepare && yarn build)
• Deploy to a staging subgraph and confirm voteStake updates on subsequent Reward/Bond/Unbond/Rebond events for active polls
• Verify that the endBlock check still correctly stops updates after a poll actually ends

🤖 Generated with Claude Code

[rickstaa]

@adamsoffer good find. Approved, as logic checks out will run some validations and then we can merge and publish a patched version.

[rickstaa]

This was validated by @adamsoffer in https://gist.github.com/adamsoffer/fb0b241203ca54f696fdeab0a5be62f9.