Add Valkey as a storage backend for CrewAI memory

Summary

Add Valkey as a storage backend for CrewAI's memory
system — a vector-search-capable store for short/long-term memory, plus a
shared cache usable by A2A and file uploads. Valkey is an open-source,
BSD-licensed fork of Redis, so this gives users a self-hostable, license-clean
alternative to the existing storage options.

Opening this to track the work per @Vidit-Ostwal's request while the repo
cleanup and the third-party-integration/configuration discussion are underway.

Status: implementation essentially complete

The work is already built, tested, and split across four open PRs (broken up at
an earlier reviewer's request for smaller, easier-to-review diffs). They are
rebased on latest main, CI is green, and CodeRabbit has passed. Best reviewed
in order:

• #5700 — shared cache config + ValkeyCache (foundation for A2A and file uploads)
• #5701 — harden memory tool input validation
• #5702 — async-safe embeddings + resilient drain_writes
• #5703 — ValkeyStorage vector memory backend (depends on #5700 and #5702)

Notes

• No changes are needed from maintainers right now — this issue is for tracking.
  I understand the third-party-provider/configuration direction is still under
  internal discussion, and I'm happy to adapt the PRs (e.g. how the backend is
  registered/configured) to fit whatever pattern you land on.
• The implementation uses Valkey-GLIDE and the Valkey Search module for vector
  similarity (FLAT/HNSW), with scope/category/metadata filtering.

The implementation is already split across PRs #5700, #5701, #5702, and #5703; review them in that order, starting with the shared cache foundation. Check the Valkey-GLIDE, Valkey Search, configuration, validation, and async-write behavior described in the issue. The work is done when the PR series is reviewed and its four listed capabilities are accepted.