fix main.py, added airgapped environment Dockerfile and added k8s resources folder

[Badlybear]

main.py:
Fixed Postgres error on file_counts updates
Replaced two assignments to the same column in one UPDATE with a single assignment (nested jsonb_set) in both single and batch endpoints.
This removes ERROR: multiple assignments to same column "file_counts" and keeps counters correct.
Made SQL result keys stable (avoid KeyError: '"id"')
Added explicit column aliases (AS id, AS content, …) in every SELECT and RETURNING.
The code now reads row["id"], row["content"], etc., reliably.
Made timestamps consistent and driver-agnostic
Converted DB timestamps to epoch seconds in SQL using EXTRACT(EPOCH FROM ...) AS ..._ts.
Removed Python .timestamp() calls (which failed when the driver returned strings).
All responses now use integer epoch fields (created_at, expires_at, last_active_at) safely.
Kept behavior but hardened pagination & limits
limit is clamped to a maximum (100).
has_more logic preserved; first_id/last_id populated from the page slice.
Safer vector search & filtering
Query embedding is computed once, cast to ::vector and used with <=>.
Metadata filters use parameterized jsonb access (metadata->>$key = $value).
Batch insert is fully parameterized
Values list constructed with positional parameters; no string interpolation of data.
Returns aliased columns for uniform response building.
Error handling & responses unchanged for API surface
Models and endpoints didn’t change shape; only internal query/row handling & stability.

Dockerfile-airgapped:
Make the image run in closed networks by baking Prisma engines into the image and pre-generating the Prisma Python client during the online build stage.
Avoid any runtime downloads; Prisma loads binaries from a local cache directory.
Keep the runtime image minimal and network-independent.