Fix two memory safety issues.

[noncrab]

I've got an application which happens to re-size the allocate lmdb mapping by closing and re-opening the database, as that happens to be easier than ensuring the DB is quiescent first in this case.

I ended up finding that my program would segmentation fault when beginning transactions, or whilst the runtime was cleaning up an exited thread. Using valgrind, I tracked down two issues:

in Database::open, we allocate an Option<CString> to hold the database name. However, Option::map consumes it's value, so any contained CString is dropped after the call to CString::as_ptr, but before we pass it to mdb_dbi_open. So we have an effective use-after-free.

I'd found that a MDB_txn was getting freed twice inside of mdb_txn_abort, and then in mdb_env_close. It turns out that an earlier call to mdb_txn_comit was failing, so the transaction had been marked as finished, but the use of the lmdb_call! macro meant we didn't zero out the TxHandle content. So, we ended up calling mdb_txn_abort on an already finalized handle. Then, because of the state of the transaction flags, we ended up freeing the memory that was referenced by the environment, as it's me_txn0 member.

Anyway, I hope this helps, and thanks for the library!

[AltSysrq]

Thanks for the PR @cstorey! I actually thought mdb_txn_commit() left the transaction open if it fails, but as you hinted it does in fact call mdb_txn_abort() implicitly.

This looks good on all points, merging.

[AltSysrq]

Uploaded 0.2.1 with this fix.