This fixes issues in tools like PhpStorm that are not able to show test progress when the test name changes.

Merged from $647

https://jira.mongodb.org/browse/PHPLIB-362

Note that I haven't added any `use` statements for new classes. This will be done automatically when we add `phpcs` checks in [PHPLIB-330](https://jira.mongodb.org/browse/PHPLIB-330).

Merged from #646

https://jira.mongodb.org/browse/PHPLIB-401

Merged from #645.

* Don't collect unexpected extra commands in spec tests: as discussed in https://github.com/mongodb/mongo-php-library/pull/642#discussion_r307108104, it makes more sense to not collect commands that we're not going to be asserting against instead of dropping them later.
* While the docs for the aggregation helpers in `MongoDB\Database` and `MongoDB\Collection` have been updated with respect to read concerns and write stages, the documentation in the `MongoDB\Operation\Aggregate` wasn't. This is done in this PR.

http://jira.mongodb.org/browse/PHPLIB-460

With these changes, we're preparing the build file to do more than just running tests. To make feedback more immediate and to save resources, the build now runs a smoke screen testing step before running the full build pipeline. This should catch simple errors and will also do static analysis before running all tests. Once smoke screen testing has completed, the full build pipeline will be executed.

This PR also adds testing against PHP 7.4 with the build currently failing due to a deprecation:
```
Function ReflectionType::__toString() is deprecated
/home/travis/build/alcaeus/mongo-php-library/tests/Operation/ExplainTest.php:17
```

This will be fixed at a later step once I've figured out what's causing this.

With this change, the build first runs a single smoke screen build against the latest versions to save on build resources. Remaining topologies, server versions and PHP versions will be tested in parallel if this first build step was successful.

Merged from #644

https://jira.mongodb.org/browse/PHPLIB-438

Merged from #643

https://jira.mongodb.org/browse/PHPLIB-413

Merged from #642.

https://jira.mongodb.org/browse/PHPLIB-402

I've extracted the `is_in_transaction` helper from `MongoDB\Collection` in a separate to functions since we need it more often. I've also decided to extract the options for the aggregate helper to a common file to avoid unnecessary duplication of option descriptions.

Inconsistent behavior for advancing the ChangeStream's key after a resume token error will be addressed by PHPLIB-456.

Added ChangeStream::getResumeToken() method (PHPLIB-435), which returns the cached resume token.

postBatchResumeToken and startAfter (introduced in PHPLIB-407) can now be used for resuming.

Replaced TailableCursorIterator with ChangeStreamIterator. In addition to avoiding getMore commands when calling rewind(), the new class also tracks the size of each cursor batch so it can capture the postBatchResumeToken from getMore commands.

Refactored the resume process such that Watch now exclusively constructs the inner iterator for ChangeStream.

UnexpectedValueException is now thrown if firstBatch and nextBatch fields are not found in aggregate and getMore responses, respectively.

Tests were revised to no longer expect rewind() to encounter an error and resume. In most tests, rewind() is a NOP because the change stream has no events to return upon creation.

Since Watch now always executes aggregate commands with APM, the check for startAtOperationTime support was moved to createResumeCallable(). We continue to only capture operationTime from the very first aggregate command.

Various assertions for current() returning null were also changed to instead check valid(). Both forms are equivalent, but checking valid() is more consistent with our iteration examples.

This can be used to ensure rewinding is a NOP (i.e. getMore will not be executed) for tailable cursors with an empty firstBatch.

This will avoid spurious test failures due to timing and ensure that each acknowledged write is returned in the next change stream iteration.