refactor(context_compressor): extract content/media strip helpers (LB4)

[andrexibiza]

Summary

God-file Feature Package (tracker #78647): agent/context_compressor.py slice LB4 — extract the content/media strip helpers into agent/context_compressor_media.py with an is-identical re-export seam.

Slice

• God-file: agent/context_compressor.py (6,883 lines @ pin 6e9cae6ac4b)
• Window: 870–1111 @ pin (446–668 banner-inclusive at LB3)
• Golden sha256: 9e0acd77719ca3f8533dba593d85e768eb8ac0aef1ae8c49021de82a50f80433 — byte-exact, verified by 5/5 blind reviewers
• Module: agent/context_compressor_media.py (9 members: _IMAGE_PART_TYPES + 8 functions)
• Deps: stdlib + agent.turn_context only — no cycle
• Seam: godfile re-exports all 9 members; cc.<m> is cm.<m> verified at runtime

Verification

• 43/43 targeted tests (seam + all prior slices + ghost-skill + redaction + zero-user-guard)
• 5/5 blind re-review APPROVED (0 CRITICAL / 0 IMPORTANT; 1 cosmetic MINOR on blank-line bookkeeping)
• ruff clean · git diff --check clean · LF-only · DCO signed

Part of #78645
Part of #78647

[andrexibiza]

Interlock: Part of #78645 + #78647. 5/5 blind APPROVED receipts in worktree tmp/PASS_*_VERDICT.md.

[andrexibiza]

Interlock completion for the context-compressor shard campaign:

• Current shard PR: refactor(context_compressor): extract durable session guards mixin (LB6) #81074
• Lock issue: Shard agent/context_compressor.py (god-file decomposition) #78645
• Campaign epic: Repo-wide godfile eradication: residual 2K tasks after #102117 #78647
• This node is part of the same coordination graph; no issue is being closed by this refactor.

The full related set is explicitly recorded in #81074, including the prior compressor shards, same-surface colliders/fixers, and their referenced issue nodes. The residual agent/context_compressor.py remains alive until the 2k Law is satisfied.

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

refactor(context_compressor): extract content/media strip helpers (LB4)

1. The mid-file # noqa: E402 import blocks are load-bearing — agent/context_compressor.py (~L9-70): imports of the new leaf modules are placed after module-level code to dodge circular imports. That ordering becomes fragile the moment a leaf module imports back from context_compressor (the sibling LB6 extraction refactor(context_compressor): extract durable session guards mixin (LB6) #81074 introduces exactly such a round-trip seam). Consider documenting the dependency direction (leaves must never import from the parent at module scope) in a comment above the block.
2. No whole-compress() E2E assertion — the seam tests verify each leaf function in isolation, but nothing proves the extracted helpers are still invoked with identical semantics end-to-end. A single compress() smoke test (real ContextCompressor, call_llm stubbed) would lock the integration; the sibling refactor(context_compressor): extract message marker helpers #81243 pattern (byte-hash) is one approach, behavior assertions are a better one.
3. HISTORICAL_TASK_HEADING moved to context_compressor_budget and re-exported through the top import — external code doing from agent.context_compressor import HISTORICAL_TASK_HEADING keeps working, but anything importing it from the module directly (tests/plugins) should be migrated to the new home to avoid a hidden two-source constant.
4. New modules have no __all__ — each leaf exports a large surface of _private names; an explicit __all__ would make the seam surface self-documenting and catch accidental new exports. (Minor.)