Memory Construction

This 10-line block is copy-pasted across SimpleMemory, WindowBufferMemory, DummyMemory:

impl Into<Arc<Mutex<dyn BaseMemory>>> for SimpleMemory {
    fn into(self) -> Arc<Mutex<dyn BaseMemory>> {
        Arc::new(Mutex::new(self))
    }
}
// repeated for DummyMemory, WindowBufferMemory

Also: Into<Arc<dyn BaseMemory>> (non-mutex) is implemented on all three but never used by any chain or executor — it is dead API.

// Will replace the 6 manual impls (3 types × 2 variants)
impl<T: BaseMemory + Send + Sync + 'static> From<T> for Arc<Mutex<dyn BaseMemory>> {
    fn from(mem: T) -> Self {
        Arc::new(Mutex::new(mem))
    }
}

Call sites using .into() will continue to work unchanged.

Until JOB-256 is merged, implement the conversions manually:

use std::sync::Arc;
use tokio::sync::Mutex;
use langchainx::schemas::memory::BaseMemory;

pub struct SlidingWindowMemory { /* ... */ }

impl BaseMemory for SlidingWindowMemory { /* ... */ }

// Required until blanket impl lands
impl Into<Arc<Mutex<dyn BaseMemory>>> for SlidingWindowMemory {
    fn into(self) -> Arc<Mutex<dyn BaseMemory>> {
        Arc::new(Mutex::new(self))
    }
}

// DO NOT add Into<Arc<dyn BaseMemory>> — it is dead API

All memory uses tokio::sync::Mutex today. Memory is read far more often than written (read on every plan() call, write only on Finish). RwLock would allow concurrent readers.

JOB-256 also evaluates switching to RwLock. Do not change this unilaterally — it is part of the JOB-256 scope.