feat(storage): use mdbx_txn_reset to time out transactions

[shekhirin]

Fixes #6441, #6699

1. mdbx_txn_reset instead of mdbx_txn_abort for timing out transactions
   • The issue with mdbx_txn_abort is that it frees the transaction pointer, meaning that MDBX can re-use it for new read-only transactions.
   • It's bad because our assumption is that they're unique and we can index the internal list of active transactions by them.
   • mdbx_txn_reset releases old snapshots of the data too, hence prevents the growth of the freelist (see https://erthink.github.io/libmdbx/usage.html#autotoc_md50).
2. Use transaction's mutex when timing out, so we're sure that if we started the process of timing out, no other parallel use of this transaction is possible
   • It's crucial, because if you try to use a transaction that was reset via mdbx_txn_reset, MDBX crashes.
3. Do not track aborted transactions in the TxnManager
   • No need for this anymore, we rely on mdbx_txn_flags(txn_ptr) & MDBX_TXN_FINISHED == 1 now.
4. Deny direct access to the transaction pointer, meaning all requests should go through the txn_execute method that does locking
   • See point 2.

[mattsse]

nice find

[emhane]

super clean interface, some test coverage missing

[emhane]

also closes #6699 right? and makes #6798 obsolete.

[emhane]

super nice! one typo in a comment (see suggestion)