LocalReadOnlyStream::stream_open refuses to open a file when 'rb' flag is passed in

This patch should solve the problem.

Writing a test is possible. So attached patch also includes a test.

Changed the test, so that only 1 handle to the file is open.

That's another error in LocalReadOnlyStream::stream_flush(): stream_flush() gets called during stream_close(). So stream_flush() should not trigger an error but fail silently.

And thus also an error in an existing test:
the test on fclose should not use @fclose as fclose is not expected to trigger warnings.

Attached are patches with corrected tests and code that removes both errors in LocalReadOnlyStream.

- LocalReadOnlyStream extends LocalStream which does implement stream_flush by returning return fflush($this->handle);. So it can't be removed.
- I did a small test: calling fclose() on a stream wrapper will first execute stream_flush() and only then stream_close(). In other words: the stream_flush is not executed by the fclose($this->handle) in the stream_close() implementation. So setting a flag in stream_close to ignore the upcoming stream_flush won't work.
- AFAIK there's no way to find out if there's something in the buffer. A bit of reading through the manual and Googling did not reveal anything that suggests that this can be found out.

So I think that leaves us no other option than to silently ignore calls to stream_flush. But as stream_write() calls are intercepted as well as any attempt to open a file other then in read-only mode, I don't think this is a problem.

By looking at the inheritance tree, I found out that ReadOnlyStream does not share its implementation with LocalReadOnlyStream and thus contains the same errors. The new patches also solve the same errors in ReadOnlyStream.

Thanks for mentioning. Changes in the attached patch:
- Accept 'rt' as mode.
- Updated method documentation as it explicitly mentioned 'r' as only acceptable mode.
- Changed the test for mode 'w' to test 'r+' instead (might someone ever try to optimize the logic...).
- Added a test for mode 'rt'

I left out the test only patch as it should be clear by now that there is a bug in the code and that we have tests that can detect it.

EDIT: cross posted, I did add that test (mentioned in #11 and have tested it on a Windows machine, Let's see if it fails on non-Windows.

This patch comes from #12 but with the logic from #11 (in_array instead of a series of comparisons).

Oops... wrong file indeed.