feat(snowflake_workspace): parameterized tasks (WITH CONFIG) + multi-instance via assets_by_name

Same three-change pattern as v0.6.5 (procs), applied to tasks:

1. _make_task_asset() gained a config_v parameter. When non-empty, the
   asset emits:
     EXECUTE TASK <db>.<schema>.<task> WITH CONFIG => '<json>'
   where <json> is json.dumps(config_v) with single quotes escaped per
   Snowflake's literal rules. Inside the task body,
   SYSTEM$GET_TASK_GRAPH_CONFIG() exposes the JSON for SQL / Snowpark
   logic. When config_v is empty, the call is the plain EXECUTE TASK
   shape (backward-compatible with v0.6.5 and earlier).

2. Factory defined once before the per-task loop so both single-instance
   and multi-instance dispatch can call it (mirrors the proc fix).

3. Task-discovery loop now reads ``assets_by_name.<task>`` directly:

   ── Single-instance (default): one task → one Dagster asset, with an
      optional ``config:`` override. Backward-compatible with all
      existing configs (no override → no config → identical
      EXECUTE TASK <name> shape).

   ── Multi-instance: one task → N Dagster assets via an ``instances:``
      list. Each instance can have its own ``config``, ``asset_name``
      (or ``key``), ``description``, ``group_name``, ``deps``, ``kinds``,
      ``tags``, ``owners``. Lets customers wire per-tenant / per-region
      task invocations as N distinct catalog assets — the parent task
      definition stays singular in Snowflake while Dagster orchestrates
      N parameterized executions.

The observation sensor's task_metadata bucket is populated for each
instance (keyed by the instance's asset_name) so completion events
land on the matching catalog node.

YAML example for multi-instance:

  assets_by_name:
    REFRESH_REGION:
      instances:
        - asset_name: refresh_us
          config: {region: "US", batch_size: 1000}
          kinds: [snowflake, daily]
        - asset_name: refresh_eu
          config: {region: "EU", batch_size: 500}
          deps: [refresh_us]

Both single + multi paths flow through the same _make_task_asset()
factory and get the v0.6.4 numeric-perf-metadata enrichment for free.
_apply_asset_overrides doesn't need touching — config: and instances:
naturally fall through its allow-list (the override is read directly
from self.assets_by_name above).

L1 validation harness scorecard unchanged:
    PASS:  562  WARN:    0  FAIL:   41

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: eric-thomas-dagster

dagster_community_components/__init__.py

@@ -21,7 +21,7 @@
 import importlib.util
 from pathlib import Path
 
-__version__ = "0.6.5"
+__version__ = "0.6.6"
 
 _PACKAGE_ROOT = Path(__file__).resolve().parent