docs(mem0): point pgvector rank-fix at upstream PR #4994

Opened the Python equivalent of the merged TS fix (#4944) at
mem0ai/mem0#4994 — update both the README
patches table and the deployment.yaml comment so the stopgap can be
dropped without re-research once #4994 merges and we bump the image.

Author: Kblack0610

apps/mem0/README.md

@@ -112,7 +112,7 @@ single `sed` patch before launching uvicorn. The patches are visible in
 
 | # | Patch | Reason | Drop when |
 |---|-------|--------|-----------|
-| 1 | `pgvector.py` line 251: `score=float(r[1])` → `score=float(1.0 - r[1])` | mem0 v2.0.1's pgvector backend returns cosine **distance** as the `score` field, but `score_and_rank` (used by `/search`) treats `score` as similarity and sorts `reverse=True`. Without this fix, the LEAST-similar memories rank first — exact-text matches end up dead last. The TS SDK was fixed in upstream PR [#4944](https://github.com/mem0ai/mem0/pull/4944); the Python equivalent is not yet upstream. | A new image tag containing the Python equivalent of #4944 lands. Then drop `command:` + `args:` from `deployment.yaml`, restoring the image's default CMD. |
+| 1 | `pgvector.py` line 251: `score=float(r[1])` → `score=float(1.0 - r[1])` | mem0 v2.0.1's pgvector backend returns cosine **distance** as the `score` field, but `score_and_rank` (used by `/search`) treats `score` as similarity and sorts `reverse=True`. Without this fix, the LEAST-similar memories rank first — exact-text matches end up dead last. The TS SDK was fixed in upstream PR [#4944](https://github.com/mem0ai/mem0/pull/4944); the Python equivalent is open as [#4994](https://github.com/mem0ai/mem0/pull/4994). | A new image tag containing #4994 lands. Then drop `command:` + `args:` from `deployment.yaml`, restoring the image's default CMD. |
 
 The container fails fast (`grep -q ... || exit 1`) if a future image bump
 silently breaks the patch (e.g., upstream rewrites the line) — better than

apps/mem0/deployment.yaml

@@ -56,10 +56,9 @@ spec:
           # but score_and_rank() treats `score` as similarity (sorts
           # reverse=True), so the LEAST-similar memories rank first. Convert
           # distance→similarity (`1.0 - r[1]`). Same fix the TypeScript SDK
-          # already received in upstream PR #4944; the Python equivalent is
-          # not yet upstream. Drop this command override + restore the
-          # image's default CMD once the Python fix lands and we bump the
-          # image tag.
+          # already received in upstream PR #4944; Python equivalent is
+          # open as PR #4994. Drop this command override + restore the
+          # image's default CMD once #4994 lands and we bump the image tag.
           # The sed range `/ORDER BY distance/,/return \[OutputData/` scopes
           # the substitution to search() only, leaving keyword_search()'s
           # ts_rank score (already-similarity) untouched. The grep gate