refactor: relocate partition iterator and expose graph builder API

Move the partition iterator implementation from obikdump to obikfilter, updating imports and exposing the type publicly. Add obikidxcache as a local dependency for obikfilter. In obikindexer, expose the new build_layer_from_kmers function to enable shared graph-building logic across pipelines without code duplication.

src/Cargo.lock

@@ -1543,6 +1543,7 @@ version = "0.1.0"
 dependencies = [
  "obicompactvec",
  "obikentropy",
+ "obikidxcache",
  "obikindex",
  "obikseq",
  "obiskio",

src/obikdump/src/dump.rs

@@ -8,7 +8,7 @@ use obikindex::KmerIndex;
 use obikidxcache::index_cache::IndexCache;
 use obikfilter::KmerFilter;
 
-use crate::partition_iter::FilteredPartitionIter;
+use obikfilter::FilteredPartitionIter;
 
 /// Raw content export of a `KmerIndex` — `KmerIndex` is a foreign type
 /// (`obikindex`), so this is an extension trait rather than an inherent `impl`.

src/obikdump/src/lib.rs

@@ -6,6 +6,3 @@
 //! reverse), same pattern as `obikindexer`/`obikquery`.
 
 mod dump;
-mod partition_iter;
-
-pub use partition_iter::FilteredPartitionIter;

src/obikfilter/Cargo.toml

@@ -10,3 +10,4 @@ obikseq       = { path = "../obikseq" }
 obiskio       = { path = "../obiskio" }
 obitaxonomy   = { path = "../obitaxonomy" }
 obikentropy   = { path = "../obikentropy" }
+obikidxcache  = { path = "../obikidxcache" }

src/obikfilter/src/lib.rs

@@ -3,19 +3,20 @@
 //! Orthogonal to genome-column selection/aggregation (`obikselect`), which
 //! operates on already-retained k-mers.
 //!
-//! [`filter`] (the `KmerFilter` trait + its implementations) depends only
+//! [`filter`] (the `KmerFilter` trait + its implementations) is the pure
-//! on `obicompactvec`/`obikseq`, not on `obikindex`. The partition/layer
+//! judging logic. [`partition_iter`] (`FilteredPartitionIter`) is the read
-//! iteration that actually runs these filters over an index
+//! side that actually runs those filters over a *complete* source index's
-//! (`iter_partition_kmers`, `iter_partition_kmers_located`) lives in
+//! partitions/layers, batched through `obikidxcache::IndexCache` for
-//! `obikdump` instead (`FilteredPartitionIter`) — it needs `obikidxcache`'s
+//! locality — used both by `obikdump` (CSV export) and by this crate's own
-//! `IndexCache` to read a *complete* source index, which this crate has no
+//! `Filter` algorithm (index-to-index rebuild).
-//! reason to depend on.
 
 mod filter;
+mod partition_iter;
 mod predicate;
 
 pub use filter::{
     GroupQuorumFilter, KmerFilter, MaxGenomeCount, MaxGenomeFraction, MaxTotalCount,
     MinComplexity, MinGenomeCount, MinGenomeFraction, MinTotalCount, passes_all,
 };
+pub use partition_iter::FilteredPartitionIter;
 pub use predicate::{GenomeSelector, GroupFilterParams, MetaPred, Selection};

src/obikfilter/src/partition_iter.rs

@@ -1,5 +1,5 @@
 //! Filtered, batch-oriented iteration over an already-cached index's
-//! partitions/layers — the read side of `obikfilter`'s `KmerFilter`s.
+//! partitions/layers — the read side of [`crate::KmerFilter`]s.
 //! `IndexCache` is a foreign type (`obikidxcache`), so this is an extension
 //! trait rather than an inherent `impl`.
 //!
@@ -13,7 +13,7 @@ use obikindex::layer::{KmerLayer, LayerContent};
 use obikidxcache::index_cache::IndexCache;
 use obikseq::CanonicalKmer;
 
-use obikfilter::{KmerFilter, passes_all};
+use crate::filter::{KmerFilter, passes_all};
 
 /// Kmers pulled per batch from a layer before filtering — keeps matrix reads
 /// grouped by (partition, layer) for locality instead of hopping row to row

src/obikindexer/src/graph_pipeline.rs

@@ -148,3 +148,24 @@ pub fn materialize_layer(
     debug!("materialize_layer: MPHF build done");
     Ok(n)
 }
+
+// ── build_layer_from_kmers ──────────────────────────────────────────────────
+
+/// Build a layer's identity (unitigs + MPHF + evidence) straight from an
+/// already-filtered kmer iterator — no file I/O, no abundance filtering:
+/// the caller has already decided which kmers survive (e.g. abundance
+/// filtering in [`crate::extensions::build_index_layer`], metadata-based
+/// filtering in `obikfilter::Filter`). Both share this same construction
+/// tail instead of duplicating the graph-building mechanism.
+pub fn build_layer_from_kmers(
+    kmers: impl Iterator<Item = CanonicalKmer>,
+    layer_dir: &Path,
+    block_bits: u8,
+    evidence: &IndexMode,
+) -> OKIResult<usize> {
+    let mut g = GraphDeBruijn::new();
+    for kmer in kmers {
+        g.push(kmer);
+    }
+    materialize_layer(g, layer_dir, block_bits, evidence)
+}

src/obikindexer/src/lib.rs

@@ -16,4 +16,4 @@ pub mod algorithms;
 pub(crate) mod extensions;
 pub mod graph_pipeline;
 
-pub use graph_pipeline::{build_graph, materialize_layer, write_graph_as_unitigs};
+pub use graph_pipeline::{build_graph, build_layer_from_kmers, materialize_layer, write_graph_as_unitigs};