Fix the bug with the wrong format.

Make the high water mark shaper better

Enable sending console output to standard error

* If the shaper is in overload and the final message comes in, but no
  further messages arrive for some time, until another message came in,
  the drop count would not be printed. Now we set a timer to ensure it
  prints the drop count after the current second expires.
* Allow the shaper to take a filter function that allows events that
  would not normally be printed anyway to not be counted against the
  HWM. This means that if you're suppressing supervisor startup messages
  you won't see drop events counted for messages you'd never see
  printed.

Fixes for OTP 20

* Handle new gen_server crash messages that contain client pid and stacktrace
* Update platform define for OTP 20
* Handle arity change in gen_fsm handle_msg

Document placeholders introduced by error logger integration

Fix path expansion for traces when log_root is set

Previously the following configuration didn't work properly:

    {log_root, "logs"},
    {handlers, [{lager_file_backend, [{file, "access.log"}, {level, none}, ...]}]},
    {traces, [{{lager_file_backend, "access.log"}, [{tag, access}], info}]}

When lager:trace_file was trying to find a file among the existing
handlers, it was searching for the expanded file name
("logs/access.log"), but the handlers store the file names without
expansion ("access.log"), so it didn't find it.

The result was that lager:trace_file started another "access.log"
backend, which of course didn't use e.g. the log rotation settings of
the first handler. When the log_root was relative, then each log
message appeared twice in the log file (both backends thought that they
should log it, because when calculating the Destination,
"logs/access.log" was used in both cases).

Now the code performs path expansion on the file name inside the handler
too. Also, now it calculates the absolute paths of both files, so that
if one file is specified with an absolute path and the other with a
relative path, lager will still know that they are the same.

Test
----

Using the old code (tag 3.2.4 was tested) the new test usually fails the
following way:

    module 'lager_trace_test'
      lager_trace_test: trace_test_ (Trace combined with log_root)...*failed*
    in function lager_trace_test:'-trace_test_/0-fun-6-'/0 (test/lager_trace_test.erl, line 71)
    **throw:{too_many_entries,{ok,<<"Test message\n2017-03-26 23:04:23.935 [info] <0.5898.0>@lager"...>>}}
      output:<<"">>

The reason for failing only usually is that it might happen that the
duplicate log entry is added to file only after we called
time:sleep(1000) and checked that the log entry is not there.

Added elixir section to README

Format gen_fsm/gen_statem crashes correctly

Add several new trace filters

Don't use dirty timeout for gen_statem

That was introduced in OTP 19.2, and causes
failures in tests on OTP 19 and 19.1

Fix statem for all 19.x releases

Add a simple case to support gen_statem crashes, since they generate the
same message in error_logger but not the same returned state.

Capture error-site metadata from error_logger messages & feed them to lager

Previously, error_logger messages only had the pid of the crashed
process, not any of the traditional lager metadata like
module, function, line, etc.

It turns out, however, that we already have most if not all of that
information as metadata already in the stack trace. Thus, we can capture
it and re-inject the error logger log message as a lager message with
metadata.

Additionally, when a process crashes, we get the process dictionary
passed to us, which we can scan for the '_lager_metadata' entry, and
attach the dead process' metadata to the log message as well.

Travis needs a higher timeout for the async test

Crude but effective fix for #393

* A side effect is you can set per processes functions via lager:md/1