* It requires a forced refresh on startup (even though that should be the
  default). Otherwise, it wouldn't print the info line correctly.
* Redirect stdin and pass it to newterm() to fix key queuing.
  Probably necessary for supporting ncurses on NetBSD as well.
* Avoid doupdate() if screen is too small: fixes crashes for very
  small windows.
* Updated Scintilla: There were some implicit typing assumptions,
  that are broken by this platform.

* After re-benchmarking the performance, I in fact detected a 20-25% speedup if
  memory limiting is active. Both using `time` and the builtin monotic timer ::^B.
  When memory limiting is disabled, there is no detectable difference to jemalloc.
  The following test case was used:
  sciteco -e '::^BUs 100000<@^U[^E\a]"^E\a" %a> ::^B-Qs='
* I also played around with getrusage(), but it doesn't seem to be a viable
  API for detecting the currently used RSS, i.e. it does not allow recovering
  from OOMs.

* We now set opt.retain=false for the process, so jemalloc returns
  freed memory and the RSS decreases when recovering from memory limit hits.
  This should be safe at least on FreeBSD.
* Either the opt.retain option is new or I was previously testing
  this only on 32-bit systems.

* teco_memory_usage is now an unsigned integer.
* Unfortunately we currently rely on the variable being int-sized since we use
  atomic operations.
  This means on 64-bit systems, limiting will not work as expected if you set the limit larger
  than 4GB.
  Not sure whether this should be fixed.
* Calling teco_memory_check() with a non-null request-size was totally broken and could
  result in bogus failures.
  This is currently used exclusively for checking backwards searches.

* There is cleanup that is not strictly necessary, because it only frees memory
  which is freed on program termination anyway.
* However, it helps to explicitly free everything for debugging memory leaks via Valgrind.
* The new macro reduces the number of #ifdef statements.
* On NDEBUG, the code of these functions will still be eliminated.
* If functions are referenced only from the destructor, there will be no unused function
  warnings, even in NDEBUG.

* On FreeBSD both the dlmalloc replacement and poll-thread via sysctl() work
  but the poll-thread has been benchmarked to be significantly faster,
  at least on my machine.
  You can still ./configure --enable-malloc-replacement of course.
* Interestingly, the RSS of the process visible via htop does not decrease
  after OOMs or command-line terminations - with neither of the implementations.