Promise catching

[brettz9]

Hi Aaron,

I've got quite a large number of changes here, but I think they are a bit interrelated (mostly on the theme of yet more comprehensive promise rejection, validation, and ensuring Firefox can now pass all tests; the bad-args.js test file gives a practical indication of what can be caught, some of which were not catchable before).

Below is a detailed summary of all of the changes in this PR. There is slightly less detail in the updated CHANGES file (avoiding mention of very minor or relatively inconsequential internal refactoring changes which I nevertheless mention below).

There were a few refactoring changes which may make it appear larger than it was though, such as switching to const/let.

Note that I tried to be conservative in extending testing timeouts for Firefox, only adding it where it was noted as necessary as per my testing results.

I was also conservative in adding try-catch blocks for the hard (but catchable) errors I could replicate, but I tried to be fairly thorough, but I may have missed some of these.

Here are the changes:

• Breaking change: Require db.cmp to use promises (so errors can be caught in analogous fashion);
• Breaking change (minor): Return error object with Promise reject calls;
• Breaking change (minor): Change bad keys error message;
• Deprecated: on schema.indexes, in place of the index key property, keyPath should be used;
• API addition: Add aliased server methods put and delete (for closer parity with spec);
• API change: Allow add/update {item:...} without key for sake of unambiguity;
• API change: Allow add/update {item:...} to be of any value including undefined or null;
• API change: Allow Mongoifying of add/update/remove keys;
• API change: Disallow key in count() if null;
• Cross-browser support: Auto-wrap user-supplied Server.error() and Server.addEventListener('error', ...) handlers with preventDefault so as to avoid hard ConstraintError aborts in Firefox;
• Cross-browser support: move rejected errors' execution from onupgradeneeded to onsuccess (Firefox);
• Cross-browser support (minor): wrap delete onblocked event's newVersion (=null) with Proxy but avoid using using Proxy if not present for sake of PhantomJS or older browsers (Firefox) - fix/improve comment wrapper for oldVersion but could not wrap it, however, so avoid testing (not part of our API anyways but do test err.type is blocked);
• Fix (error checking): Reject (and break) upon errors (createObjectStore, bad index creation, bad function on modify() object, bad modify result, bad map() function, and bad put()), and test for such bad argument promise catching as well as testing regular try-catching for event listener methods;
• Fix: Ensure limit() arguments are numeric (and test);
• Fix: Add in missing (though already documented) chaining of (short) event handlers;
• Fix: Allow empty string indexes;
• Fix (data validation): Tighter checking on argument to modify method (ensure is an object);
• Fix (data validation): Better createIndex validating (issue #149)
• Testing (Gruntfile): Add "phantom" task (separate task but still run by npm test); avoid displaying Saucelabs or adding testJobs when not needed (e.g., for "dev"); document
• Testing: get Firefox to avoid errors by lengthening test timeouts (adding as needed)
• Docs: Clarify db.delete() "success" on non-existent database, "schema" property, modify
• Document add/update's key/item properties and flattening behavior
• Refactoring: Reorder Server to be after IndexQuery (since invoking it) and put Server.query at top so it can be closer to IndexQuery which it calls
• Refactoring: Use const/let where possible (and make consistent and grouped together)
• Refactoring: Exit loop upon Promise rejection for add()/update();
• Refactoring: Comment non-throwing code to indicate why certain potentially throwing code is not being wrapped in try-catch
• Refactoring (minor): complete simplifying of object keys with same name as variable in scope, capitalize comments, iterate using Object.keys to avoid need for hasOwn checks
• Refactoring (minor): Move key conversion to top of function (so can exit earlier if problematic);
• Refactoring (minor): Remove unused second argument accidentally supplied to resolve

[brettz9]

I know there is a lot here, but can I ask when u think u may find some time to review? Thanks!