feat(cli): resolve PostgreSQL settings from app.yaml, not just env

[eldonm]

Summary

create_server_from_config threads mongodb's uri/name and dynamodb's table/region into jvspatial's DatabaseConfig — but had nothing for postgres. The DSN reached the driver only through jvspatial's own env read, so an app.yaml database.uri was silently ignored for postgres while working for every other backend.

This wires the four postgres_* settings through, and bumps the pin to jvspatial==0.0.17.

Related issues

Depends on TrueSelph/jvspatial#37, released in 0.0.17 (now on PyPI). Completes the Postgres work from #139 / #140.

Type of change

• Bug fix (non-breaking)
• New feature (non-breaking) — app.yaml now configures Postgres
• Breaking change
• Documentation
• Chore / tooling / CI — pin bump

Changes

File	Change
jvagent/cli/server_config.py	database.uri / pooler_mode / min_pool_size / max_pool_size → DatabaseConfig for postgres/postgresql
tests/cli/test_server_config_postgres.py	new — 7 tests
pyproject.toml	pin 0.0.16 → 0.0.17
docs/postgres.md	caveat replaced with a worked app.yaml example
.env.example, environment-keys-reference.md	notes corrected
.planning/reference/jvspatial-integration.md	pin refs §1 and §5
CHANGELOG.md	[Unreleased] entry

Checklist

• I read CONTRIBUTING.md and the local jvagent/cli/CLAUDE.md.
• pre-commit run --all-files passes.
• pytest tests/ passes — 0 failures, run against the released 0.0.17 wheel.
• Bug fixes cite file:line in the description.
• Docs / CHANGELOG.md updated.
• No ADR was modified in place.

Notes for reviewers

Why the pin has to move. Before 0.0.17, DatabaseConfig discarded anything passed by field name — its aliased fields lacked populate_by_name, so DatabaseConfig(postgres_dsn=...) silently produced None. This passthrough would have been written correctly and done nothing. That is in fact how the upstream bug was found: the code looked right and the tests failed. Worth knowing because the same shape was also silently breaking DynamoDB's app.yaml config, for as long as those fields have existed.

Shared database.uri. Postgres reads the same app.yaml key as mongodb, with database.type deciding interpretation. The alternative was a database.postgres_uri, which would have made the stanza backend-specific in a way the existing keys are not. Easy to change if you'd rather they were distinct.

A typo shouldn't stop the server booting. A non-integer pool size is logged and dropped rather than raised, so the driver default applies. Covered by test_non_integer_pool_size_is_ignored.

Verified against the released wheel, not a checkout. pip install jvspatial==0.0.17 from PyPI: full suite 0 failures, and scripts/smoke_postgres.sh 15/15 — bootstrap, serve, JWT login, agent turn, server restart, and recall of pre-restart state from Postgres.

One process note: my first verification of this branch failed, because uv pip install -e with a relative --python path installed into the sibling repo's venv instead of this one, and import jvspatial; print(jvspatial.__file__) reported the source path anyway due to cwd shadowing. If you verify locally, check the specific module — python -c "import jvspatial.api.config_groups as cg; print(cg.__file__)" — which is unambiguous.

Steps to test

pytest tests/cli/test_server_config_postgres.py -q
scripts/smoke_postgres.sh          # spins up its own container

For the new path specifically, an app.yaml with no JVSPATIAL_POSTGRES_* env set:

config:
  database:
    type: postgres
    uri: postgresql://user:pw@localhost:5432/mydb
    pooler_mode: transaction