Tracking issue for PBT deletion tests

[kevaundray]

Since the PBT introduces the concept of code deduplication, it can have interesting edge cases wrt deletion and 7702.

This is a tracking issue for interesting deletion cases, we might want to consider

[kevaundray]

Taken from a chat with AI(not readable):

Live code replacement reclaims old chunks

Build a diff that replaces a sole-holder account's non-empty contract code with different non-empty code, and assert the incremental state root equals a fresh embed of the post-state.

Restores coverage from the deleted test_code_change_swaps_the_chunk_leaves. It will fail today, so it forces an explicit decision: either mark it xfail with a note that the diff is fork-unreachable, or restore the reclamation arm in apply_diff_to_trie (state_pbt.py:278-292) that this PR removed. Without it the orphaning is silent.

Deleting the last holder of multi-group code drops all groups

Delete the sole holder of code spanning multiple code groups and assert every chunk group is removed and the root matches a fresh post-state embed.

Restores coverage from the deleted test_code_change_by_the_last_holder_drops_every_group, now framed as a deletion. The PR's rewrite (test_deleting_the_last_holder_drops_every_group in test_state_pbt.py) may already cover this; confirm it asserts multi-group removal before adding.

Deployment to occupied address aborts

Fork-level test that CREATE or CREATE2 targeting an address with existing non-empty code is rejected (EIP-7610 collision).
Pins the "deployment requires an empty code hash" invariant that apply_diff_to_trie (state_pbt.py:214-219) now silently depends on. If a future fork weakened this, the chunk-reclamation removal would become an actual bug, and this test would catch it at the source.

EF-prefixed deploy code is rejected

Fork-level test that init-code returning bytes beginning 0xEF is rejected by EIP-3541 (the 0xEF-prefix check in interpreter.py).
Pins the "an indicator may not be deployed as code" invariant. Complements the existing test_contract_hashing_to_the_delegation_marker_executes_as_code in test_code_sharing.py, which covers the hash-collision half but not deployment rejection of the literal prefix.

Live code change on a non-delegation asserts

Add an assertion in apply_diff_to_trie (state_pbt.py:285-292) that a live account's non-empty pre-state code hash is either unchanged or a delegation, then test that a diff violating it raises.

Converts the current silent orphaning into a loud failure. Makes the implicit invariant explicit at the exact point the code now relies on it, so any future fork feature that mutates deployed contract code fails here instead of committing a corrupt tree.

[kevaundray]

Also adding tests around https://eips.ethereum.org/EIPS/eip-7610