Avoid race between numreaders++ and reading numreaders at cleanup. Make
the un-mutexed reset of reader table entry, atomic: Reset mr_pid only.

Instead check mr_pid != 0 in mdb_page_alloc()'s scan for readers.
(txnid_t)-1 as "no ID"-mark avoids a check for mr_txnid != 0.
The scan can stop when seeing an old reader.

An open MDB environment does not survive or catch fork(),
so repeating getpid() was pointless.

Rename USE_POSIX_SEM to MDB_USE_POSIX_SEM.
Separate MDB_FDATASYNC from MDB_USE_POSIX_SEM.

Fix typos, error code ranges

Allow cursors on read-only txns to be reused with later txns.

Overwrites the passed in key with the DB's key

More for d7935941 ITS#7229

Note read-write vs read-only mmap tradeoffs.

Fix abort of child txn

due to a0993354

Otherwise a process that opens and closes an env more than once
will leak mutex handles. (WIN32/BSD)

This is quick and dirty, should be integrated into mdb_stat.

Ignore current cursor pos if it's out of bounds.

Minimizes the number of memcpy's if additional dups are stored.

Append mode now does no key comparisons, input must be in sorted order.
page_split was not updating cursor parents correctly.

Just lookup the DB roots as needed. When many DBs are in use,
most of the copies won't be referenced in a given txn, and
there's a bad race condition in the copy routine.

Must begin with '/'