Fix random crashes and subsequent test failures in umbrella apps

There is a possibility of a race condition on stopping/restarting
ExUnit/AssertValue apps more frequent within umbrella projects.
When ExUnit is stopped, AssertValue may still attempt to flush the
StringIO. This may lead to various errors.
Changed code flushing ExUnit's IO to handle these states.

Also changed AssertValue.Server to always restart on errors
(default GenServer behaviour)

lib/assert_value/server.ex

@@ -1,5 +1,5 @@
 defmodule AssertValue.Server do
-  use GenServer, restart: :temporary
+  use GenServer
   import AssertValue.StringTools
 
   def start_link(args) do
@@ -65,8 +65,26 @@ defmodule AssertValue.Server do
   end
 
   def handle_cast({:flush_ex_unit_io}, state) do
-    contents = StringIO.flush(state.captured_ex_unit_io_pid)
-    if contents != "", do: IO.write(contents)
+    # There is a possibility of a race condition occurring when
+    # stopping/restarting ExUnit/AssertValue applications within umbrella apps.
+    #
+    # When ExUnit is stopped, AssertValue may still attempt to flush the
+    # StringIO. There are two potential states to consider:
+    #
+    # 1. if captured_ex_unit_io_pid is nil, this will result in
+    #    "no function clause matching in StringIO.flush/1."
+    # 2. if captured_ex_unit_io_pid is not nil, but the process is stopped,
+    #    this will result in "(EXIT) no process: the process is not alive..."
+    #
+    # Code below handles both of these cases
+    try do
+      if state.captured_ex_unit_io_pid do
+        contents = StringIO.flush(state.captured_ex_unit_io_pid)
+        if contents != "", do: IO.write(contents)
+      end
+    catch
+      :exit, {:noproc, _} -> :ok
+    end
     {:noreply, state}
   end