Add support of NOT REGEXP operator to PostgreSQL + fix testRegexCondition

vaplas created an issue. See original summary.

The test is based on testRegexCondition(), only the results are replaced by their opposites + used no-deprecated assert methods.

Also nit improve pattern test method testRegexCondition().

-    $query->condition('t.age', '2[6]', 'REGEXP');
-    $result = $query->execute()->fetchField();
-    $this->assertEquals($result, '26', 'Regexp with number type.');
+    $query->condition('t.age', '2[5]', 'REGEXP');
+    $result = $query->execute()->fetchCol();
+    $this->assertEquals($result, [25], 'Regexp with number type.');

@daffie, your comment makes sense to me, definitely!

IMHO, we have 3 in 1 tests :) And only one of them is sensitive to the operator (REGEXP or NOT REGEXP). Two other relate to the processing of the regular expression as a whole, and are independent of the operator.
Such:

// Ensure that filter by "#" still works due to the quoting.
It has depend only from 'pattern' of regular expression (via SQLite reason).

// Ensure that REGEXP filter still works with no-string type field.
It has depend only from 'type of field' of regular expression (via PostgreSQL reason).

Also, during refactoring, I discovered one magical piece:

+  public function testNotRegexCondition() {
...
+    $test_groups[] = [
+      'regex' => 'Ringo|George',
+      'expected' => [
+        'George',
+        'Paul',
+      ],
+    ];

This found George. What is he doing here? Patronage by Foreach? After refactoring such a feint will not work :)

Interdiff not attached, because it does not seem helpful.

Patronage by Foreach?

Not. By sort(), because it return only true|false. Hence:

$this->assertEqual(count($result), count($test_group['expected']), 'Returns the expected number of rows.');
$this->assertEqual(sort($result), sort($test_group['expected']), 'Returns the expected rows.');

always pass with equal sizes.

Bit cleanup.

Super nice, I can look at the #8 patch all day long!)

Also I was wrong when I thought that case with 'no-string' does not depend on the operator. We really need to check both operators. Sorry for the noise about this.

After #8 we lost few comments:

// Ensure that filter by "#" still works due to the quoting.
...
// Ensure that REGEXP filter still works with no-string type field.

But maybe the git logs are the best way to analyze the reason for the appearance of these cases in providerRegularExpressionCondition.

RTBC for me. But it will be great if someone will confirm this too.

+++ b/core/tests/Drupal/KernelTests/Core/Database/SelectTest.php
@@ -443,42 +443,40 @@ public function testRandomOrder() {
+    return[

nit typo return [.

I will not cry much on these comments :) I just wanted to say, that I saw it during the review and do not mind, because the git-history is better tool for this (I read about it in book "Clean Code / Robert Martin"). But I also do not mind saving comments, of course.

So, we can not just duplicate the current REGEXP test to NOT REGEXP, because this leads to a number of problems:

• ugly copy-paste,
• illegibility (echoes of simpletest),
• duplication sort() error.

Hence the changes in scope.

#8 patch good fix problems with database support regular expression (complete pair REGEXP - NOT REGEXP, and correction tests). Just fix nit typo.

Update IS and title, because we also fix the error in the testRegexCondition.