Optimisations relating to get_expected_withdrawals (#9314)

Tweaking/optimising `get_expected_withdrawals` on the pre-Gloas path, after the removal of some optimisations in:

- https://github.com/sigp/lighthouse/pull/9102


  - Remove `prepare_beacon_proposer` calls from `register_validator`. These calls were: heavily duplicated, happening _too early_ (at the start of the slot before the block arrived), expensive on epoch boundaries (forcing an early epoch transition), _and_ unnecessary. They are completely safe to remove because the scheduled proposer preparation routine exists.
- Instrument `get_expected_withdrawals`/`prepare_beacon_proposer` to find more unexpected callers.


Co-Authored-By: Michael Sproul <michael@sigmaprime.io>

Co-Authored-By: Tan Chee Keong <tanck@sigmaprime.io>

Co-Authored-By: Michael Sproul <michaelsproul@users.noreply.github.com>

beacon_node/execution_layer/src/lib.rs

@@ -406,6 +406,10 @@ impl ProposerPreparationDataEntry {
     }
 }
 
+// NOTE: This key should arguably include the `suggested_fee_recipient`, as it is part of the
+// `Proposer::payload_attributes` value that is cached based on it. However, in some cases where
+// this key is constructed the fee recipient is not straight-forward to determine. Therefore we
+// accept the risk of loading a stale fee recipient within the timespan of a single slot.
 #[derive(Hash, PartialEq, Eq)]
 pub struct ProposerKey {
     slot: Slot,