Reverts #9829 (Third attempt at merging #9809)

Okay. This change failed on staging twice in a row, but has passed CircleCI and local tests every time. Something must be different about the staging box. Starting an investigation...

Here's the failure from the build log (repeated several times - it retried, I think?)

FAILED TESTS:
  SentByDropdown
    – only shows each name once
      PhantomJS 2.1.1 (Linux 0.0.0)
    expected { Object (component, root, ...) } to have a length of 1 but got 0
    AssertionError@[CDO]/apps/test/unit-tests.js:470:25 <- webpack:///~/chai/~/assertion-error/index.js:74:0
    assert@[CDO]/apps/test/unit-tests.js:4571:32 <- webpack:///~/chai/lib/chai/assertion.js:107:0
    assertLength@[CDO]/apps/test/unit-tests.js:5646:17 <- webpack:///~/chai/lib/chai/core/assertions.js:1045:0
    assert@[CDO]/apps/test/unit-tests.js:4370:55 <- webpack:///~/chai/lib/chai/utils/addChainableMethod.js:84:0
    [CDO]/apps/test/unit-tests.js:261874:89 <- webpack:///test/unit/netsim/NetSimLogBrowserFilters.js:81:67

Here's the source of that test, along with everything I think is relevant:

// ./apps/test/unit/netsim/NetSimLogBrowserFilters.js
import React from 'react';
import {shallow, mount} from 'enzyme';
import {spy} from 'sinon';
import {expect} from '../../util/configuredChai';
import {throwOnConsoleErrors} from '../../util/testUtils';
import NetSimLogBrowserFilters, {SentByDropdown} from '@cdo/apps/netsim/NetSimLogBrowserFilters';
import i18n from '@cdo/netsim/locale';

describe('SentByDropdown', function () { // Line 55
  it('only shows each name once', function () {   // Line 73
    const result = mountWithLogRows([
      {'sent-by': 'Alice'},
      {'sent-by': 'Alice'},
      {'sent-by': 'Bob'},
      {'sent-by': 'Bob'},
      {'sent-by': 'Bob'}
    ]);
    expect(result.find('option[value="by Alice"]')).to.have.length(1); // Line 81, the failed assertion
    expect(result.find('option[value="by Bob"]')).to.have.length(1);
  });

function mountWithLogRows(logRows) { // Line 107
    return mount(
      <SentByDropdown
        i18n={i18n}
        currentSentByFilter="none"
        setSentByFilter={spy()}
        logRows={logRows}
      />
    );
  }
});

Checking for environmental differences between my machine and staging:

Suspect #1 is enzyme, which is important to this test and is off by a minor version. Installing enzyme 2.3.0 on my local machine and running grunt unitTest reproduces the failure!

Pinning enzyme to 2.4.1 will fix the problem, but I'd like to understand what changed and why my test is failing. The enzyme 2.4.0 changelog contains some fixes to attribute selectors; sure enough, rewriting my assertions to avoid attribute selectors gets them passing under 2.3.0:

// expect(result.find('option[value="by Alice"]')).to.have.length(1);
expect(result.find('option').findWhere(n => n.prop('value') === "by Alice")).to.have.length(1);

What bothers me is that I can't figure out which change actually fixed this selector. enzymejs/enzyme#412 and enzymejs/enzyme#387 look most relevant but it's not clear that either should have directly affected the selector option[value="by Alice"] which freaks me out a bit. Update: Found it! This PR fixes a bug with spaces in selectors, and was not listed in the changelog for some reason. I'll contact enzyme and suggest this get mentioned.

In any case, I'd prefer not to switch to this; CSS-style attribute selectors are supposedly supported by Enzyme and they do work under 2.4.1. Instead we're going to pin to enzyme 2.4.1 which supports selectors with spaces, to avoid differences between dev and staging in the future.