random_ber_signature only emits long form lengths in signatures it marks as certainly_not_der, so run_ecdsa_der_parse asserts that such inputs are rejected but never that a well-formed long form length is accepted. As a result, inverting any of the four conditions in secp256k1_der_read_len that govern the long form goes unnoticed by the test suite.
Add the accepting cases, together with the two rejecting ones that share their boundaries: a length below 128, which must use the short form, and length octets with a leading zero. Since DER permits the long form only for lengths of at least 128, which a signature with two in-range scalars never reaches, the R integers in the accepting cases necessarily overflow to zero; the test checks the parsed values so this stays pinned down.
I noticed this lack in the tests while working on mutation testing for the ecdsa impl. So these tests kill the following mutants: https://secp256k1.space/src/ecdsa_impl.h#863, https://secp256k1.space/src/ecdsa_impl.h#878 and https://secp256k1.space/src/ecdsa_impl.h#896
