Revert #845 lock hardening and #865 zip/nginx changes (write_lock worker failures on k8s/NFS)

[brianlball]

Why

Recurring write_lock worker failures on the k8s/NFS cluster running 3.10.0-179 images (which carry #845 + the #865-equivalent #863). Bisect branch 179D-310 (= develop@Jul16 + #852 + #853, i.e. develop minus #845/#865, on OS 3.10) ran clean on the cluster (3.10.0-179D-test images). This PR makes develop match the known-good configuration.

Mechanism (theory consistent with sim logs): on NFS, flock state is server-authoritative but File.exist? is subject to attribute/negative-dentry caching, so #845's lock_abandoned? probe (2 probes 3s apart) falsely declares the lock holder dead during the receipt-visibility lag, deletes the live lock, fails datapoints, and triggers a re-extract herd over a live analysis dir.

What

• Revert 40d2bfd (Fix analysis-zip corruption on shared-FS (k8s/NFS) deployments (#857); nginx keepalive/gzip (#864) #865: read-only zip extract initialize_worker zip lock uses flock which is not cross-host safe on NFS, causing zlib corruption race on Kubernetes #857, verified download, receipt re-check, extract_archive ignores overwrite parameter, causing stale measure.xml files to persist #858 overwrite, nginx keepalive/gzip nginx: keepalive_timeout 0 and gzip off degrade Cloud Management Console load performance #864)
• Revert 821fcf7 (Harden RunSimulateDataPoint worker against permanent lock deadlock and stale-job re-dispatch #845: lock_abandoned? probing, fail-fast, stale-job re-dispatch guards)
• Cherry-pick 9b93a5b: widen morris/sobol algorithm-results-zip size caps (200k→230k, 40k→46k) — full rserve rebuilds on push runs drift past the old caps (seen on 179D-310 CI)

Kept intact: #852 (BSON cap / terminal state) and #853 (nil-DataPoint guards — verified 3× if d.nil? in resque job). The #845/#853 hardening specs are deleted with the revert (they test #845 code).

Known trade-offs

• Reintroduces the rubyzip read-path rewrite issue (initialize_worker zip lock uses flock which is not cross-host safe on NFS, causing zlib corruption race on Kubernetes #857) and drops the extract_archive ignores overwrite parameter, causing stale measure.xml files to persist #858 overwrite honor + nginx keepalive 65/gzip (nginx: keepalive_timeout 0 and gzip off degrade Cloud Management Console load performance #864). Those fixes should return later in an NFS-safe form (e.g. lookupcache=positive mount opt, probe persistence > cache staleness, inode revalidation after flock).

🤖 Generated with Claude Code