[Bug] Search panics on disk-write failure and crashes the entire all-in-one process #3796

Describe the bug

When a Bleve/zapx segment persist fails (disk quota exceeded), search panics with assignment to entry in nil map in invertedIndexCache. On all-in-one that kills the whole process (search, proxy, storage-users, NATS, postprocessing), exit 2, CrashLoopBackOff.

Steps to reproduce

1. Run all-in-one OpenCloud with Bleve v5 / zapx v17.
2. Hit a write failure on bleve-v5/store/*.zap (disk quota exceeded) during an index batch.
3. Have a search query in flight at the same time.

Expected behavior

A failed persist is an error in search. The rest of the process keeps running.

Actual behavior

{"level":"error","service":"search","error":"segment: /var/lib/opencloud/search/bleve-v5/store/000000003df6.zap persist err: write /var/lib/opencloud/search/bleve-v5/store/000000003df6.zap: disk quota exceeded","time":"2026-10-22T13:39:47Z","line":"github.com/opencloud-eu/opencloud/services/search/pkg/search/service.go:690","message":"error adding updating the resource in the index"}

Then:

panic: assignment to entry in nil map

github.com/blevesearch/zapx/v17.(*invertedIndexCache).insertLOCKED(...)
        github.com/blevesearch/zapx/v17@v17.2.3/inverted_text_cache.go:94

Two goroutines hit the same nil map — search query (SearchInContext → TermFieldReader) and index write (Scorch.Batch → DocNumbers). Proxy had been serving (GET /graph/v1.0/me → 200); Kubernetes then saw :9200 connection refused. df still showed free space; the write error was disk quota exceeded.

Setup

All-in-one Helm, 1 replica, Recreate. Data PVC CephFS RWX 300 Gi at /var/lib/opencloud.

Image=docker.io/opencloudeu/opencloud-rolling:8.1.0
STORAGE_USERS_DRIVER=decomposed
STORAGE_SYSTEM_DRIVER=decomposed
SEARCH_EXTRACTOR_TYPE=tika
OC_ASYNC_UPLOADS=true

Additional context

bleve/v2@v2.6.1, zapx/v17@v17.2.3, reva/v2@v2.51.0.