Problem

Several internal caches in OpenCode use standard Map objects without size limits. Over long-running sessions, these caches can grow without bound, contributing to memory issues like those reported in #5363.

Affected Caches

1. Instance cache (project/instance.ts) - Caches project instances by directory
2. Provider SDK cache (provider/provider.ts) - Caches loaded SDK instances
3. LSP spawning cache (lsp/index.ts) - Tracks LSP server spawn promises

Code Example

// Current: unbounded Map that grows forever
const cache = new Map<string, Context>()

// No eviction when cache gets large
cache.set(directory, context)

Impact

• Memory usage grows continuously over time
• Long-running sessions can consume excessive RAM (70GB+ reported in opencode eating 70gb of memory? #5363)
• No mechanism to evict stale entries
• Opening many directories causes unbounded growth

Relates to #5363

Expected Behavior

Caches should have bounded size with LRU eviction:

const cache = createLruCache<string, Context>({
  maxEntries: 20,
  onEvict: (key, value) => {
    // Clean up evicted entry
    value.dispose()
  }
})

Proposed Solution

Create a bounded LRU (Least Recently Used) cache utility that:

• Limits maximum entries with configurable maxEntries
• Automatically evicts least-recently-used entries when limit is reached
• Provides onEvict callback for cleanup when entries are removed
• Maintains Map-like interface for drop-in replacement