Race condition in MDBX bindings

[shekhirin]

Describe the bug

With the introduction of transaction manager, we now have three places where the transaction can be aborted:

1. Drop implementation when it's a read-only transaction

   reth/crates/storage/libmdbx-rs/src/transaction.rs

   Line 365 in 9d9d7ee

   ffi::mdbx_txn_abort(txn);

2. Message listener in TxnManager

   reth/crates/storage/libmdbx-rs/src/txn_manager.rs

   Line 98 in 9d9d7ee

   sender.send(mdbx_result(unsafe { ffi::mdbx_txn_abort(tx.0) })).unwrap();

3. Read transactions monitor in TxnManager

   reth/crates/storage/libmdbx-rs/src/txn_manager.rs

   Line 249 in 9d9d7ee

   let result = mdbx_result(unsafe { ffi::mdbx_txn_abort(ptr) });

Recently, we saw double free or corruption (!prev) and Segmentation fault (core dumped) errors coming from the FFI boundary. Most likely, it's related to the race condition happening somewhere between read transactions monitor and the user-invoked transaction abort (either Drop or message listener).

1. Segmentation fault is most likely caused by the attempt to use a transaction that was already aborted
2. Double free is most likely caused by the attempt to abort the transaction twice