Fix bug in Markov.Chain where reply_chance would get reset, add regression tests for this

* Markov.Chain.add_weight/4 would return a fresh new markov chain,
  instead of using all of the previous values of the chain that was
  being modified. This would result in resetting the reply_chance value
  to the default of 0.01.
* Add tests to make sure this doesn't happen to add_weights in the
  future.

Signed-off-by: Alek Ratzloff <alekratz@gmail.com>

test/contrib/markov/chain_test.exs

@@ -68,6 +68,23 @@ defmodule MarkovChainTest do
     }
   end
 
+  test "chain add_weight does not reset reply_chance" do
+    chain = %Chain {order: 2, reply_chance: 0.0}
+            |> Chain.add_weight(["foo", "bar"], "baz")
+
+    chain = chain |> Chain.add_weight(["foo", "bar"], "baz", 2)
+    assert chain.reply_chance == 0.0
+
+    chain = chain |> Chain.add_weight(["foo", "bar"], "qux")
+    assert chain.reply_chance == 0.0
+
+    chain = chain |> Chain.add_weight(["bar", "baz"], "qux")
+    assert chain.reply_chance == 0.0
+
+    chain = chain |> Chain.add_weight(["bar", "baz"], nil)
+    assert chain.reply_chance == 0.0
+  end
+
   test "chain merge works correctly" do
     chain1 = %Chain {order: 2}
             |> Chain.add_weight(["foo", "bar"], "baz")