feat(opencode): add double-buffer context management

[marklubin]

Issue for this PR

Refs #8140 #11314 #2945 #4317 #12479

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Implements double-buffer context management — a two-phase compaction strategy that produces higher-quality summaries by running summarization early (at ~50% capacity) while the model still has full attention, rather than at the capacity cliff.

How it works:

1. Checkpoint phase (~50%): A background compaction generates a summary and stores it in per-session state. The user's workflow is uninterrupted.
2. Concurrent phase (50-75%): Normal operation continues. The pre-computed checkpoint summary is ready.
3. Swap phase (~75%): Instead of stop-the-world compaction, uses the pre-computed summary. Falls back to standard compaction if no checkpoint is available.

Changes:

• double-buffer.ts: State machine tracking normal → checkpoint_pending → concurrent → swap phases per session
• compaction.ts: Four new exports — checkpointIfNeeded() (fire-and-forget background checkpoint), shouldSwapBuffer(), getCheckpointSummary(), completeSwap()
• processor.ts: Skip overflow on compaction summaries (fixes Bug: opencode run exits after compaction when compaction model's token usage exceeds overflow threshold #13946, also submitted separately as fix(opencode): skip overflow check on compaction summaries #15129)
• prompt.ts: Swap check before standard overflow + background checkpoint trigger after successful processing
• config.ts: Configurable checkpointThreshold and swapThreshold in compaction schema

Community asks addressed:

• Configurable compaction thresholds (Feature Request: Configurable context limit and auto-compaction threshold #8140, Feature Request: Configurable Context Compaction Threshold #11314)
• Background compaction that doesn't block the user (Feature: generic /compact command, auto-compaction, and fork-aware conversations #4317)
• Higher-quality summaries by summarizing earlier (Session automatically compacted, destroying the entire working context #2945, [FEATURE]:Session compaction includes already-summarized messages, wasting tokens #12479)

Existing single-pass compaction remains as the fallback when no checkpoint is available.

Algorithm description: https://marklubin.me/posts/hopping-context-windows/

How did you verify your code works?

• Unit tests for state machine transitions (double-buffer.test.ts)
• All 109 session tests pass
• Full test suite: 1157/1158 pass (1 pre-existing unrelated failure)

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Found related PRs! These are related to the compaction and context management work but address different aspects:

Related PRs (Not Duplicates):

1. PR fix(opencode): skip overflow check on compaction summaries #15129 - fix(opencode): skip overflow check on compaction summaries

   • This is a companion fix referenced directly in PR feat(opencode): add double-buffer context management #15130 (mentioned in the description)

2. PR feat(compaction): configurable threshold and model (global + per-model) #11931 - feat(compaction): configurable threshold and model (global + per-model)

   • Related to configurable compaction thresholds mentioned as one of the community asks

3. PR feat(session): add custom compaction thresholds along with prevention of continuous compaction #10123 - feat(session): add custom compaction thresholds along with prevention of continuous compaction

   • Addresses configurable compaction thresholds (Feature Request: Configurable context limit and auto-compaction threshold #8140, Feature Request: Configurable Context Compaction Threshold #11314)

4. PR fix(opencode): prevent context overflow during compaction #11453 - fix(opencode): prevent context overflow during compaction

   • Related to overflow prevention during compaction

5. PR fix: recover from 413 Request Entity Too Large via auto-compaction #14707 - fix: recover from 413 Request Entity Too Large via auto-compaction

   • Related to automatic compaction strategies

6. PR feat(opencode): add context overflow prevention with configurable thresholds #8810 - feat(opencode): add context overflow prevention with configurable thresholds

   • Addresses similar threshold configuration needs

These are all related work in the compaction/context management space but none are duplicates of PR #15130. This PR is the first to implement the specific "double-buffer" two-phase compaction strategy with checkpoint phases.