[FIX] increment analysis order on POST instead of always 1

[lobennett]

Summary

Fixes #1580. Creating an analysis via POST /api/analyses/ without an explicit order always assigned order = 1, so analyses never received a sequential order within their study.

Root cause

AnalysisSchema.load_values (a @pre_load hook) auto-assigns the order by querying the current max for the study:

if data.get("study_id") is not None:
    max_order = db.session.query(func.max(Analysis.order)).filter_by(study_id=data["study_id"]).scalar()
    data["order"] = 1 if max_order is None else max_order + 1
else:
    data["order"] = 1

The field is declared study_id = fields.String(data_key="study"). @pre_load runs on the raw payload, before marshmallow maps the data_key, so the study id is present under the key study, not study_id. data.get("study_id") was therefore always None, the else branch always ran, and every new analysis got order = 1.

Fix

Resolve the id from the study data_key (keeping the study_id fallback and a nested-object guard), mirroring the pattern PointSchema.process_values already uses for the identical case:

study_id = data.get("study_id") or (
    data.get("study") if isinstance(data.get("study"), str) else None
)
if study_id:
    max_order = db.session.query(func.max(Analysis.order)).filter_by(study_id=study_id).scalar()
    data["order"] = 1 if max_order is None else max_order + 1
else:
    data["order"] = 1

AnalysisSchema is the only change; PointSchema already handled this correctly.

Tests

The existing test_post_analysis_without_order only asserted order is not None, which passed even with the bug (order was 1, not None) — that gap let the bug through. Adds test_post_analyses_without_order_increments_within_study: creates a fresh study and POSTs two order-less analyses, asserting orders 1 then 2. Fails on the old code (both 1), passes with the fix.

python -m pytest neurostore/tests/api/test_analyses.py -k order
# 3 passed