fix: conserve staged-removal validator balance on deferred top-up

[matthias-wright]

Builds on #276 (which builds on #275).

Addresses #374.

Changes:

• Guard the Inactive deposit branch on account.balance > 0 to distinguish a staged-removal validator (still holds a credited balance) from a new-validator placeholder (balance 0). For the former, withdraw the full credited balance (balance_deduction = balance), refund the top-up separately, and let the completion path remove the account. Placeholders and active top-ups are unaffected.
• Add regression test test_last_block_topup_does_not_drop_staged_removal_balance.

[sebastian-osec]

Question: should the top-up refund in this staged-removal path be taxed as an invalid deposit? The top-up passed validation when accepted; it is refunded only because the earlier staged removal wins. Since this branch uses push_invalid_deposit_withdrawals, a nonzero invalid_deposit_tax sends part of the valid top-up to treasury. If that is intended policy, this is fine, but the regression test only covers tax = 0.

[matthias-wright]

Good question. I think a tax-free withdrawal is more appropriate here.

Update(ca07b67):

• Refund staged-removal top-up untaxed

[sebastian-osec]

Thanks. If we are okay with max_withdrawals_per_epoch applied once to validator withdrawals and once to refund withdrawals, then my concern is resolved. #276 .

One small follow-up: the docs still describe max_withdrawals_per_epoch as limiting total withdrawals in the last block of each epoch, not as a per-type limit:

summit/docs/deposits-and-withdrawals.md

Line 123 in ca07b67

The `max_withdrawals_per_epoch` protocol parameter (range: 1–256) limits how many withdrawals can be included in the last block of each epoch. If more withdrawals are scheduled for an epoch than the cap allows, only the first `max_withdrawals_per_epoch` withdrawals (in queue order) are included. The remaining overflow withdrawals are rescheduled to the next epoch, placed at the **front** of that epoch's queue so they have priority over withdrawals originally scheduled for that epoch.

The latest commit addresses my original tax concern too. The staged-removal top-up refund now bypasses the invalid-deposit tax helper and refunds the full top-up amount, with a nonzero-tax regression test covering it.