feat: support bind_all in topology/update-conf and rocksdb store tuning fields in update-conf

[thweetkomputer]

Summary

EloqKV only listens on the node's configured ip unless bind_all is set (the gflags default is false), so eloqctl-deployed nodes were unreachable via any interface other than the deployment address — e.g. 127.0.0.1 refuses connections when a node is deployed with an internal IP. eloqctl had no way to set this key.

Also completes #431: the RocksDB store tuning knobs were configurable via launch YAML but not via update-conf.

Changes

• Add optional deployment.bind_all topology field (default false, matching the eloqkv built-in default); written to the [local] section of the generated Eloqkv.ini, following the same pattern as enable_io_uring.
• Add bind_all=${OVERRIDE} to the EloqKv.ini template.
• Whitelist bind_all for eloqctl update-conf --fields so existing clusters can enable it without redeploying.
• Whitelist rocksdb_periodic_compaction_seconds and rocksdb_delete_obsolete_files_period_micros for update-conf --fields. EloqKV reads these from the [store] section, but update-conf inserted missing keys into [local] unconditionally — new keys are now routed to the section the field belongs to (created when absent).
• Ignore the test-generated src/cluster_mgr/upload/ directory.

Testing

• cargo build / cargo build --release
• cargo test --lib: all pass (note: test_s3_endpoint_written_to_store_and_txlog_ini is flaky when run in parallel with config::monitor tests — pre-existing env-var race on UPLOAD_PATH_DIR, unrelated to this change; passes in isolation)
• Manually verified on a local cluster that the generated node ini gets bind_all=false by default and bind_all=true when set in topology

🤖 Generated with Claude Code