fix: include jinja2 as a core llama-stack dependency (#1529)

We removed `llama-models` as a dep which was pulling this in for us
previously. This did not get caught in the release process because the
distros we use for testing (fireworks / together) pull that in via
sentence transformers which we don't use in all distros (notably
ollama.)

See #1511 

## Test Plan

Ran `llama-stack-ops/actions/test-and-cut/main.sh` with
`ONLY_TEST_DONT_CUT=1 COMMIT_ID=origin/fix_jinja2` and by making it
build the ollama docker. Ran the docker to ensure it does not error out
with jinja2 dependency error. (Unfortunately there is another error with
sqlite_vec there.)

pyproject.toml

@@ -25,6 +25,7 @@ dependencies = [
     "fire",
     "httpx",
     "huggingface-hub",
+    "jinja2>=3.1.6",
     "jsonschema",
     "llama-stack-client>=0.1.6",
     "prompt-toolkit",