reward automation gating paths have no tests (zero / two delegations, not-an-assigner)

[gyorgybalazsi]

Summary

Three gating decisions in the reward automation have no test at any level, and none is reachable from the localnet IT, which only exercises the happy one-delegation / node-is-an-assigner path:

Path	Location	Behaviour
zero active delegations	reward_automation.rs active_delegation	silent no-op
more than one active delegation	active_delegation	returns Err, every tick
this node's member is not an assigner	run_once_for_party	log and skip

The middle one is the consequential case: it is the tick-level half of the total-loss mode this PR guards at the propose boundary (two live delegations). If it ever stopped returning Err and started guessing which split was authoritative, that would be rewards paid to a superseded split — and nothing would catch the regression.

Why they are untested

active_delegation performs the ACS read and the ambiguity decision in one function, so there is no seam to test the decision without a ledger. The same is true of the assigner check inside run_once_for_party.

Requirements

1. Separate the decision from the I/O: a pure function taking the decoded records (or a count) and returning no-op / act / refuse, and a pure check for "is this node's member an assigner". parse_delegation_record is already factored out this way and unit-tested, so the pattern exists in the file.
2. Unit-test all three outcomes, with the two-delegations case asserting that it refuses rather than picks — the assertion should read as a statement about safety, not about a return type.
3. Leave the ledger reads where they are; this is about testability of the rules, not restructuring the reads.

References

• crates/decman/src/server/reward_automation.rs — active_delegation (ambiguity refusal, zero-delegation no-op), run_once_for_party (assigner skip).
• crates/decman/tests/common/phases/coupon_reassignment.rs — the IT, happy path only.