Clear displayed_etag on etag-less full refresh

[balloob]

Problem (MJ-17)

handleDirectWriteEnd in src/display_service.cpp commits displayed_etag only inside a hasNewEtag && newEtag != 0 guard. An etag-less full upload (or an auto-complete END with no etag payload) changes the panel pixels but leaves displayed_etag at its previous value.

The partial-update path (0x76) validates with:

if (oldEtag == 0 || oldEtag != displayed_etag || newEtag == 0) { /* ETAG mismatch -> full fallback */ }

So a later partial from another client whose oldEtag happens to equal the now-stale displayed_etag passes this check and diffs against the wrong, outdated base image — silent visual corruption.

Fix

On a successful refresh where no valid etag was supplied, clear displayed_etag to 0 (invalid/unknown) instead of leaving a stale value. 0 is never a legitimate displayed_etag match (a partial with oldEtag == 0 is already rejected), so the next partial gets a clean ETAG mismatch and correctly falls back to a full upload. No bogus etag is invented; a later client simply re-uploads in full.

Behavior when a real etag IS supplied is unchanged, as is the partial-refresh commit path (displayed_etag = partialCtx.new_etag on success).

State transitions

1. Etag-less full upload succeeds → displayed_etag = 0.
2. Next partial (0x76) arrives with any oldEtag → oldEtag != displayed_etag (0) → clean ETAG mismatch NACK.
3. Client falls back to a full upload (supplying a real etag if it has one) → base image is correct again.

Previously step 1 left the old etag in place, so step 2 could false-match and diff against stale pixels.

Verification

Builds pass:

• pio run -e nrf52840custom — SUCCESS
• pio run -e esp32-s3-N16R8 — SUCCESS

🤖 Generated with Claude Code

https://claude.ai/code/session_012g2e8mr132vcizx92WsgiR