[zbyyehyehyeh]

GETDEL is currently covered by a single case that uses .Val(), and
go-redis returns an empty string for both a missing key and an empty value,
so the two behaviours are not distinguished.

This adds coverage for:

• a missing key returning nil instead of an empty string;
• an empty string value, which is still distinguishable from a missing key
  and must delete the key;
• a key that carries a TTL;
• a non-string value being rejected with WRONGTYPE.

All assertions pass against unstable locally:

./x.py test go build -- -run 'TestString' ./unit/type/strings

AI assistance

Produced with qwen3.7max. I reviewed every line and ran the tests locally.

Why this matters

This is a test-only change with no behaviour change in src/. The existing
case cannot tell a missing key apart from an empty value, because go-redis
collapses both into "", so a regression in GETDEL handling of either path
would go unnoticed. The new assertions use Result()/Exists() so the two
outcomes are actually distinguishable, and cover the TTL and wrong-type paths
that were previously untested.

[github-actions]

Hi @zbyyehyehyeh,

Thank you for your pull request. Please review our Contributing Guide.

Please make sure you understand your changes and explain your reasoning in this pull request. Low-quality pull requests may be closed.

[zbyyehyehyeh]

Responding to the checklist above.

• What the change does: adds a test-only subtest for GETDEL covering four
  cases — a missing key (must reply nil, not an empty string), an empty string
  value (must still delete the key), a key carrying a TTL, and a non-string
  value (must fail with WRONGTYPE). No production code is touched.
• Why: the existing case uses .Val(), and go-redis maps both a nil reply
  and an empty bulk string to "", so the current test cannot tell the two
  apart — a regression on either path would go unnoticed.
• How I verified it: ./x.py test go build -- -run 'TestString' ./unit/type/strings
  against unstable locally; the new subtests pass under both server
  configurations used by the suite.
• AI usage: qwen3.7max helped draft the wording and the subtest skeleton.
  I chose the four assertions, checked each against Redis semantics, and ran
  the tests myself.