Replaced this with the managed AsnDecoder, removing PInvoke from a potential hot path.

Also removed the manual API calls to ldap_create_sort_control - this is now built in managed code.
This then has knock-on effects to eliminate the SortKeyInterop classes.

Most of the Control tests were hardcoded to the output of BerConverter, which uses four-byte lengths in all cases.
This behaviour is now different: the same output is returned across all platforms for .NET, and remains unchanged for .NET Framework.
This should also close issue 34679.

Reduce number of copies required in TransformControls, and enable these copies to take advantage of newer intrinsics where available.

Windows domain controllers may return a distinguished name starting with OU=, rather than ou=.

Preallocating space for AsnWriter buffers to reduce memory usage.
Correctly handling attribute names in SortControls.

Corrected off-by-one error in stack allocation comparison.
Made some stackallocs constant (unless a meaningful constant size would be >256 bytes) and reviewed these constant values based on real-world attribute lengths/server names.

* Added limited caching of AsnWriters, as small/medium/large buckets.
* Replaced several Debug.Asserts with BerExceptions.
* Changed the type of raised exception to retain backwards-compatibility.
* Added extra checks on BER content lengths to match BerConverter.

DirectoryControl payload parsing is stricter, throwing exception when invalid sequences are discovered instead of optional OCTET STRINGs.

* Ensured that all references to DirectoryControl names in comments are accurate.
* Specified the name of the ASN.1 structure which is being parsed in the comments.
* Added explanatory comment stating that BER INTEGER values are being interpreted as 32-bit signed integers.

* Refactored LDAP string writing logic into a separate WriteLdapString method in AsnWriterExtensions.
* Replaced separate AsnWriter instances with a single one (reduced memory usage when compared to multiple instances being active, even with varying buffer sizes.)
* Resolved backwards-compatibility issue when attempting to process an attribute name which contains invalid UTF-8 characters.
* Correcting expected ASN tag when parsing an attribute name in a sort result.