Evaluate Trader results with bench craft potential

[mcagnion]

Fixes #5437.

Description of the problem being solved:

Trader ranks items as they are listed on the trade site. A rare item with an open prefix or suffix, or an item with a replaceable crafted modifier, can therefore appear below a less promising item even when a simple bench craft would make it the better upgrade.

This adds an opt-in Bench Craft option to Trade Query Options. For generated queries, PoB adds an exactly-one-empty-affix pseudo-stat weighted from the compatible bench craft with the highest estimated query weight for the item type. For each fetched item, it fully evaluates the highest-level craft in each compatible local group and at most one positive highest-estimated compatible non-local craft, then selects the best calculated outcome from that bounded set. It does not exhaustively compare every compatible craft, so modifier interactions and breakpoints can occasionally make another non-local craft better. Pasted queries, or results whose selected weights changed after the search, evaluate all highest-level legal craft candidates. Result evaluation yields between build calculations so that fallback remains responsive.

The selected craft can fill an open prefix or suffix, or replace exactly one existing crafted modifier when the calculated result is an improvement. Items with multiple crafted modifiers are only considered for additions; their existing crafts are never replaced.

The result tooltip shows the selected craft or replacement, and holding Ctrl previews the simulated item with the new lines marked [Craft]. The generated weighted query uses the empty-affix pseudo-stat to influence which listings the Trade API returns; the post-fetch evaluation then reranks those listings. The simulated craft never replaces the original listing text used for import.

Steps taken to verify a working solution:

• Added focused coverage for affix identity and capacity, item-type restrictions, modifier conflicts, single-crafted-mod replacements, multi-line crafts, Catalyst modifiers, multiple crafted modifier limits, corrupted and mirrored items, immutable query-weight snapshots, bounded candidate selection, exact fallback behavior, preview rolls, tooltip lifecycle, and preservation of the original result string used for import.
• Ran the full automated test suite.
• Compared the same fetched listings with Bench Craft disabled and enabled, verified the changed order, tested a crafted-mod replacement, and checked the exact Ctrl preview in the UI.
• Replayed ten fixed listings on a saved complex build after one warm-up. Across five runs, the exhaustive fallback used 1,230 build calculations and a median of 4.98 s (4.87–5.10 s); the generated-query path used 20 calculations and a median of 0.67 s (0.65–0.69 s). Network time was excluded.

Link to a build that showcases this PR:

No specific build is required. The behavior can be reproduced with an equipment search that returns a rare item with an open affix or one replaceable crafted modifier.

Before screenshot:

After screenshot — open affix:

After screenshot — crafted modifier replacement:

Query Options:

Notes:

• The calculation is opt-in and disabled by default.
• Generated queries fully evaluate the highest-level craft in each compatible local group plus at most one positive highest-estimated non-local craft per item. This bounded selection deliberately trades exhaustive certainty for fewer build calculations.
• Pasted queries and stale weight snapshots evaluate all highest-level legal craft candidates. This path may take longer, but advances cooperatively without blocking the interface.