bump version to 0.15.2 (#278)

Fix tag codec packaging mismatch (#248)

use litellm range

fix: wire compact_semantic_duplicates setting to env var (#182)

* fix: wire compact_semantic_duplicates setting to env var

the compact_semantic_duplicates parameter in compact_long_term_memories()
was hardcoded to True with no corresponding Settings field. setting
COMPACT_SEMANTIC_DUPLICATES=false as an env var had no effect because
pydantic-settings never read it. add the field to Settings so the
docket perpetual scheduler respects the env var.

* fix: resolve compact_semantic_duplicates default at call time

the default was evaluated at function definition time, meaning
test-time patches and runtime config reloads had no effect.
use None sentinel with runtime resolution instead.

add regression test for COMPACT_SEMANTIC_DUPLICATES env var parsing.

* style: run pre-commit format to pass ci linter

make format uses a different ruff version than the pre-commit hook
(v0.3.2). re-ran with pre-commit run --all-files to match CI.

Merge pull request #138 from redis/fix/docket-task-timeouts

fix: Add per-task timeouts for LLM-dependent Docket tasks

Merge pull request #138 from redis/fix/docket-task-timeouts

fix: Add per-task timeouts for LLM-dependent Docket tasks

Merge pull request #138 from redis/fix/docket-task-timeouts

fix: Add per-task timeouts for LLM-dependent Docket tasks

Fix empty text causing OpenAI embeddings API error (#134)

## Problem
The compact_long_term_memories background task failed with:
litellm.exceptions.BadRequestError: OpenAIException - Error code: 400 -
{'error': {'message': "'$.input' is invalid..."}}

This occurred because a memory with empty text was passed to OpenAI's embedding API,
which rejects empty strings.

## Solution
Two-layer approach:

1. **Reject empty text at API layer** - Prevent new memories with empty text from being created:
   - Added validation to CreateMemoryRecordRequest
   - Added validation to UpdateWorkingMemory
   - Added validation to LenientMemoryRecord

2. **Filter legacy data gracefully** - Handle existing invalid memories in the database:
   - Filter out memories with empty text/id in index_long_term_memories
   - Skip semantic deduplication for empty text in deduplicate_by_semantic_search
   - Raise clear ValueError in embed_documents/aembed_documents for empty strings

## Changes
- agent_memory_server/models.py: Add validation to request models
- agent_memory_server/long_term_memory.py: Filter invalid memories during indexing
- agent_memory_server/llm/embeddings.py: Validate inputs before calling OpenAI
- tests/integration/test_deduplication_e2e.py: Tests for both API rejection and legacy handling

Bump version to 0.13.1

fix(ci): use NPM_TOKEN for npm publish