fix(spicepod): JSON schema accepts string or {name: expr} for partition_by (spiceai#10352)

* fix(spicepod): schema now accepts string or {name: expr} for partition_by

The custom partition_by deserializer accepts either a plain expression string
(e.g. "bucket(10, user_id)") or a single-entry object mapping a name to an
expression (e.g. {year: "YEAR(created_at)"}), but the generated JSON schema
derived from the PartitionedBy struct only accepted objects with explicit
'name' and 'expression' fields. This caused editors (yaml-schema) to report
'Incorrect type. Expected PartitionedBy' for valid spicepod configs.

Add a schema-only PartitionedBySchema helper describing both accepted shapes
and apply schemars(with = ...) on the acceleration and vector partition_by
fields, then regenerate .schema/spicepod.schema.json.

* fix(spicepod): schema accepts serde aliases (datasets.mode, eviction_policy)

Serde '#[serde(alias = "...")]' attributes are not reflected in the JSON
schema generated by schemars, so spicepod configs that used an accepted
alias (e.g. 'datasets[].mode' instead of 'datasets[].access') were flagged
as unknown properties by editors when the containing struct uses
'deny_unknown_fields'.

Add a small post-processing pass in the spicepodschema tool that walks the
generated schema and, for each known canonical field, injects an alias
property pointing at the same schema value. Currently covers:
- datasets[].mode -> datasets[].access
- eviction_policy -> caching_policy (cache configs)

Regenerate .schema/spicepod.schema.json.

* fix(spicepod): constrain partition_by named mapping to single entry

Address review feedback: the Named variant of PartitionedBySchema now sets
minProperties=1 / maxProperties=1 so editors flag invalid multi-entry
mappings instead of silently accepting them.

* fix(spicepod): update acceleration parameters to sort and deduplicate accelerator names

* Update docs comments for partition_by, to be correctly generated in
spicepod json schema

* remove schema-only descriptions

* keep only valida partition_by examples

* unify partition_by commebts

* fix(spicepod): address PR review feedback for partition_by schema

- Tighten deserialize_partition_by to error on multi-entry maps,
  non-string values, and unsupported scalar items instead of silently
  accepting them
- Add serialize_with = "serialize_partition_by" to VectorStore.partition_by
  so serialization matches the accepted YAML shapes
- Update partition_by doc comments on Acceleration and VectorStore to
  describe both accepted shapes (string | single-entry map)
- Sort and dedup data connectors and accelerators at collection time for
  deterministic schema output (removes duplicates like duckdb/arrow and
  stabilizes x-spice-connectors ordering)
- Add unit tests for inject_field_aliases covering match, mismatch,
  no-overwrite, recursive injection, and no-op cases
- Add negative tests for deserialize_partition_by rejecting invalid shapes
- Regenerate .schema/spicepod.schema.json

* refactor(tests): simplify error handling and improve readability in partition_by tests

* refactor(collector): improve sorting and deduplication logic for connector registrations

---------

Co-authored-by: ewgenius <hey@ewgenius.me>

Author: lukekim