Fix light client merkle proofs (#7007)

Fix a regression introduced in this PR:

- https://github.com/sigp/lighthouse/pull/6361

We were indexing into the `MerkleTree` with raw generalized indices, which was incorrect and triggering `debug_assert` failures, as described here:

- https://github.com/sigp/lighthouse/issues/7005


  - Convert `generalized_index` to the correct leaf index prior to proof generation.
- Add sanity checks on indices used in `BeaconState::generate_proof`.
- Remove debug asserts from `MerkleTree::generate_proof` in favour of actual errors. This would have caught the bug earlier.
- Refactor the EF tests so that the merkle validity tests are actually run. They were misconfigured in a way that resulted in them running silently with 0 test cases, and the `check_all_files_accessed.py` script still had an ignore that covered the test files, so this omission wasn't detected.

testing/ef_tests/tests/tests.rs

@@ -955,13 +955,9 @@ fn kzg_recover_cells_and_proofs() {
 }
 
 #[test]
-fn beacon_state_merkle_proof_validity() {
-    BeaconStateMerkleProofValidityHandler::<MainnetEthSpec>::default().run();
-}
-
-#[test]
-fn beacon_block_body_merkle_proof_validity() {
-    BeaconBlockBodyMerkleProofValidityHandler::<MainnetEthSpec>::default().run();
+fn light_client_merkle_proof_validity() {
+    MerkleProofValidityHandler::<MinimalEthSpec>::default().run();
+    MerkleProofValidityHandler::<MainnetEthSpec>::default().run();
 }
 
 #[test]