Seam pool stores duplicate samples of the same capture

Found during the joint-selection work (PR #154): the stored pool holds five samples that are only three distinct fits — two are exact duplicates, the same file pooled on two viewings. A capture watched twice is double-counted in every pooled statistic.

The pool should recognize a returning capture (same file identity) and refresh its sample rather than append a twin.

Out of scope for #154. Filed under the owner's standing authorization (2026-08-05: "Always feel free to file issues to keep track").

Search the code for the seam pool logic (where samples are inserted and pooled statistics are updated) and identify how capture identity is compared. Reproduce the issue by running through a capture twice and confirming the same file appears as duplicate pool entries with doubled stats. Then inspect any existing tests around pool insertion/statging or add coverage there so the expected behavior is that a returning capture replaces/refreshes its sample instead of appending one.

由索引模型根據 Issue 內容生成。