Don't remap MBEDTLS_ERR_MPS_RETRY to MBEDTLS_ERR_SSL_WANT_READ

[hanno-becker]

MBEDTLS_ERR_MPS_RETRY signals that the last API call should be retried. It's different from MBEDTLS_ERR_SSL_WANT_READ in that is does not imply that more data is required from the underlying transport.

In mbedtls_ssl_mps_remap_error(), we're remapping MBEDTLS_ERR_MPS_RETRY to MBEDTLS_ERR_SSL_WANT_READ, which looks like a temporary patch rather than something thought through.

This issue is to track finding and implementing a proper solution to this. Two approaches are:

1. Loop at some point within the TLS stack when MBEDTLS_ERR_MPS_RETRY is generated. That's what the legacy messaging stack does when the equivalent internal error code MBEDTLS_ERR_SSL_CONTINUE_PROCESSING is generated.
2. Forward the error code to the user. In this case, it probably still needs to be remapped to the SSL namespace, added to the public API, and documented.