Allow incremental together with clustered

[Dtenwolde]

Follow-up to #32, where we temporarily disallowed using incremental=true and cluster_terms=true together.

This PR allows both options to be used together.

Compared to #32 clustered incremental maintenance has some extra cost because we preserve ordered terms access across inserts and deletes, but the overhead stays modest.

Benchmark summary:

• Small datasets: essentially parity, generally within noise / ~10%.
• 100k dataset:
  • Inserts: roughly equal to non-clustered incremental.
  • Mixed workloads: roughly equal to non-clustered incremental.
  • Deletes: largest overhead.
    • Batch delete of 100 documents: ~1.6x slower.
    • Batch delete of 500 documents: ~1.3x slower.
    • Other delete cases: generally within ~1.1x slowdown.

Overall, clustered + incremental index maintenance performs similarly for inserts and mixed workloads, with some delete overhead, but all measured scenarios stayed below 2x slowdown.

[Dtenwolde]

I suppose running CI in reduced mode is fine?

[lnkuiper]

Won't this ORDER BY here make performance worse? The incrementally appended data (after the initial index creation) is not physically clustered, which I am OK with, but I don't think forcing an ORDER BY will improve performance