There are two variants of ECDSA: randomized or deterministic. The calculation of a signature involves a number $k$ which must not ever be repeated for distinct messages with the same key. There are two ways to implement that: generating $k$ randomly, or generating $k$ in a deterministic way from the key and the message. (I am omitting details here, just focusing on what's relevant for this question.) The two variants produce compatible signatures: it's impossible to tell which variant was used to produce a given signature. (Of course you can tell by asking an implementation to sign the same message twice and comparing the outputs.)
The original definition of ECDSA only considered the randomized variant, and this is the only one that NIST allows, as of FIPS 186-4 (see §6.3 and §6.4). RFC 6979 is a specification for the deterministic variant.
NIST test vectors were generated using randomized ECDSA. A consequence of using a randomized algorithm is that you cannot have known-answer tests for it. Every time you run a signature calculation, you get a different result. All you can do for testing is to test that the output of the signature calculation is a valid signature, not that it's a particular sequence of bytes.
mbedtls_ecdsa_sign_det_ext
implements deterministic ECDSA. You can use known-answer tests for deterministic ECDSA. But NIST doesn't provide test vectors for it since NIST doesn't standardize deterministic ECDSA.
If you stretch this a bit, you could say that it's possible to have known-answer tests if you fully specify how the algorithm consumes the output of the random generator. But then as a theoretical matter, you're now testing a deterministic algorithm that's taking some input which in the real world is supposed to come from a random generator. And as a practical matter, in an actual implementation, it isn't always possible to swap out the random generator for testing. And in any case, FIPS 184-4 allows some variation in how the output of the random generator is consumed when generating $k$ for an ECDSA signature.