refactor(pipeline): PR#7 slice 2 -- extract extraction.rs (issue #67 hotspot split)

Wave 1 slice 2 (docs/plans/2026-08-19-evidence-architecture-execution-plan.md
Part E): move-only extraction of per-file parsing/resolution (Tier-0/1/2/3)
and its DB-persistence counterpart out of pipeline.rs into
pipeline/extraction.rs. Moved: FORMAL_RESOLUTION_TIMEOUTS,
formal_resolution_timeout_count, formally_resolved_names, extract_file_data
(~480 lines), chunk_pending, persist_file.

formal_resolution_timeout_count (1 real external caller, calm-server's
tools/recover.rs) is re-exported at its unchanged crate::indexer::pipeline
path. extract_file_data/persist_file are pub(super) so pipeline.rs's own
reindex_* callers keep working unchanged. formally_resolved_names is also
pub(super) + imported behind #[cfg(test)] since two tests still living in
pipeline.rs call it directly. ExtractedFile/CallSiteData stay defined in
pipeline.rs (shared by other not-yet-extracted slices) and are pulled in via
`use super::{...}`.

Also trimmed now-unused imports from pipeline.rs's top-level use block
(chunk_file, insert_symbols_batch/insert_import_edges_batch, and 6 of 8
parser:: functions) since their only callers moved to extraction.rs;
CodeChunk/ParsedSymbol stay imported in both files since ExtractedFile (not
moved) still needs them.

Zero net logic diff: every function body is byte-identical to its pre-move
form, confirmed via diff against the read taken immediately before the move.

Verified: cargo build + clippy (calm-core, --all-features, --all-targets)
clean, zero warnings. Full test suite 1306 passed/0 failed.
golden_graph_equivalence's continued-vs-fresh mutation-round tests both pass.

diff_impact flags "medium" risk because every moved symbol looks "new" at its
new file path to static call-graph analysis -- expected false-positive for a
pure file move, mitigated by the build/clippy/test/golden-equivalence
verification above.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: Your Name