Explain: Improve phrasing around the last-value tool's output.

We don't need to echo the "Now" part of the tool's output when it occurs, it's
better to write something more descriptive.  The prompting can drop explanations
of the output format, which is now composed specifically for the LLM's benefit.

Author: mark-undoio

explain/explain.py

@@ -355,9 +355,9 @@ def _call_handler(_):
 
         message = result_forwards.output
 
-        if m := re.search(r".*Now =.*", message):
+        if m := re.search(r".*Now = (.*)", message):
             # If we found a value change, extract just the "Now =" part of the message.
-            message = m[0]
+            message = f"has just been assigned value {m[1]}"
 
         # Put short messages on the same line as "Expression changed".
         separator = " " if len(message.splitlines()) == 1 else "\n"

explain/instructions.md

@@ -93,10 +93,6 @@ source lines:
 This works even when the memory location was modified a long time ago or in a
 different function.
 
-When `last` returns it specifies a "Was" value, which is what you were
-investigating.  "Now" is the previous value, which should be ignored or used
-for a new debugging hypothesis.
-
 `last` will only work for values that are currently in scope.  If a value is not
 currently in scope then you need to navigate backwards in time until it is.  Using
 `reverse_step`, `reverse_next` and `reverse_finish` can help you work