fix: compress uploads for streaming-decompression-only configs; rename ZIPXL

[g4bri3lDev]

Summary

transmission_modes bit 0x01 has been repurposed over time (RAW → ZIPXL → streaming decompression), and post-2.0 device configs may set only that bit, without TRANSMISSION_MODE_ZIP (0x02). The compression gate only checked supports_zip, so such devices silently uploaded every frame uncompressed — found live on a Seeed EN05 (transmission_modes=0x09): 48 KB raw per frame instead of ~400 bytes.

• upload_image/_dispatch_upload now compress when either bit is set
• API renamed to the current semantics, with back-compat aliases kept:
  • DisplayConfig.supports_streaming_decompression (primary) — supports_zipxl/supports_raw remain as deprecated aliases
  • FIRMWARE_ZLIB_WINDOW_BITS (primary; it is the firmware inflate limit, applying to all compressed uploads since fix: always compress image uploads with a 9-bit zlib window (C3) #96) — ZIPXL_ZLIB_WINDOW_BITS remains as alias
  • CLI transmission label ZIPXL → STREAMING
• Docstrings document both config conventions (pre-2.0 devices: ZIP-only; post-2.0: streaming-only) and the 50 KB cap rationale (whole-blob buffer of the old buffered decompressor; not applied for streaming)

Hardware verification (EN05 OD5A2F4C, 800×480 MONO, panel 0x0027, via BLE DFU downgrades)

config \ firmware	1.70 (buffered uzlib)	1.73 (streaming @15-bit)	2.0 (streaming @9-bit)
ZIP-only (pre-2.0 convention)	✅ compressed accepted	✅ compressed accepted	✅ compressed accepted
streaming-only (post-2.0 convention)	⚠️ START NACKed → uncompressed fallback works	⚠️ START NACKed → uncompressed fallback works	✅ compressed accepted

The NACKs on old firmware come from its ZIP-bit gate at compressed START (removed in fw 1.83, 62f03db), not from the 9-bit zlib window — every firmware generation accepts the 9-bit stream, answering the pre-2.0 compatibility question about #96.

Test plan

• new regression test: transmission_modes=0x09 device uses the compressed protocol
• pytest -q → 494 passed; ruff, mypy clean
• hardware matrix above; C4 chunked write_config and fix: preserve real config data through JSON export/import (M7) #104 JSON round-trip also exercised live during the config swaps

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[g4bri3lDev]

Added a second gate fix found during hardware testing: the partial stream (0x76) compression decision also checked only supports_zip, so streaming-only devices sent partial streams uncompressed — measured 96,000 bytes on the wire for a full-frame partial that compresses to well under 1 KB. Same one-line fix as the upload gates.