refactor error handling part 1 (#307)

- created `getErrorMessage` function which takes in an error string or any type of error object and returns the correct error message
- replaced `error.message` with `getErrorMessage` since `error` could be a string in some cases
- merged sendNotificationForError.ts with ErrorHandlingUtils.ts
- some minor refactoring

In part 2, I will make the following changes:
 - Make `Logger.logError` function take an error message string instead of an error object. This will reduce some redundancy where the `getErrorMessage` function is being called twice (the error object passed by the caller is already an error message).
 - Update every `TelemetryProcessor.traceFailure` call to make sure we pass in an error message instead of an error object since we stringify the data we send.

src/Common/CosmosClient.ts

@@ -1,6 +1,7 @@
 import * as Cosmos from "@azure/cosmos";
 import { RequestInfo, setAuthorizationTokenHeaderUsingMasterKey } from "@azure/cosmos";
 import { configContext, Platform } from "../ConfigContext";
+import { getErrorMessage } from "./ErrorHandlingUtils";
 import { logConsoleError } from "../Utils/NotificationConsoleUtils";
 import { EmulatorMasterKey, HttpHeaders } from "./Constants";
 import { userContext } from "../UserContext";
@@ -69,7 +70,7 @@ export async function getTokenFromAuthService(verb: string, resourceType: string
     const result = JSON.parse(await response.json());
     return result;
   } catch (error) {
-    logConsoleError(`Failed to get authorization headers for ${resourceType}: ${error.message}`);
+    logConsoleError(`Failed to get authorization headers for ${resourceType}: ${getErrorMessage(error)}`);
     return Promise.reject(error);
   }
 }