Handle PoisonError when locking session

The Racer.session mutex can become poisoned if a racer call panics since
this mutex is guaranteed to be held at that time. The Session type
shouldn't panic while modifying its internal state, and it should be
safe to recover from the PoisonError that occurs.

src/engine/racer.rs

@@ -75,7 +75,10 @@ impl<'a> SemanticEngine for Racer<'a> {
 
     fn find_definition(&self, ctx: &Context) -> Result<Option<Definition>> {
         let (query_src, pos, path) = self.build_racer_args(ctx);
-        let session = self.session.lock().unwrap();
+        let session = match self.session.lock() {
+            Ok(guard) => guard,
+            Err(poisoned) => poisoned.into_inner()
+        };
 
         // TODO catch_panic: apparently this can panic! in a string operation. Something about pos
         // not landing on a character boundary.
@@ -108,7 +111,10 @@ impl<'a> SemanticEngine for Racer<'a> {
 
     fn list_completions(&self, ctx: &Context) -> Result<Option<Vec<Completion>>> {
         let (query_src, pos, path) = self.build_racer_args(ctx);
-        let session = self.session.lock().unwrap();
+        let session = match self.session.lock() {
+            Ok(guard) => guard,
+            Err(poisoned) => poisoned.into_inner()
+        };
 
         let matches = ::racer::core::complete_from_file(query_src, path, pos, *session);