TRSSBackend repeated OOM crash-loop — streaming build log offset drift causes unbounded memory growth

Summary
---

TRSSBackend has been OOM-crashing repeatedly (`FATAL ERROR: ... JavaScript heap out of memory, SIGABRT`), even with a 4GB heap already configured via `NODE_OPTIONS=--max_old_space_size=4096` in the init script. Confirmed 7+ crashes over a ~24h window on trssrtp1.

Every crash is preceded (within minutes) by a `BuildProcessor: processBuild(): Exception: Output size X > size limit 52428800` warning, with X ranging from ~58MB to ~150MB.

Root cause seems to be

In `BuildProcessor.execute()`, the "Streaming" branch (used while a build is still running) computes the next fetch offset like this:

```
startPtr = result[0].output.length;
```

Long-running, verbose builds (z/OS, AIX, JCK) are the most likely to both exceed 12MB while streaming and later show up as "output size > 50MB" in the finished-build warning — consistent with what's in the logs.

Locate BuildProcessor.execute() and inspect the Streaming branch, starting with how result[0].output and startPtr are used across successive fetches. Trace the offset and output-size handling against the reported 12MB and 50MB limits, then verify that long-running verbose builds no longer cause repeated output growth or OOM crashes.

Written by the indexing model from the issue text.