fix: avoid linear objectHas scans for large inline imports

[He-Pin]

Motivation:
PR #1061 was closed against the obsolete fixjson base, but the issue still needs a clean PR on the current upstream master. PR #1060 added the InlineScanMax fallback for valueRaw, yet repeated std.objectHas / std.objectHasAll checks over large imported strict-JSON objects still reached containsVisibleKey / containsKey and scanned uncapped inline field arrays. That kept membership-heavy workloads at O(N^2).

Modification:

• Rebase the PR fix: avoid linear objectHas scans for large inline imports #1061 follow-up onto current upstream/master (1db39dde).
• Apply Obj.InlineScanMax to Val.Obj.containsKey and Val.Obj.containsVisibleKey, so large inline objects fall through to the existing lazy getValue0 LinkedHashMap path.
• Reuse Obj.InlineScanMax in inline-object producers, keeping construction and lookup thresholds on one source of truth.
• Add a benchmark regression covering repeated std.objectHas / std.objectHasAll calls over imported JSON object fields.

Result:

Benchmark	Current master	This branch	Change
repeated std.objectHas / std.objectHasAll over large imported JSON object fields	97.942 ms/op	21.117 ms/op	4.64x faster (-78%)

Large inline imported objects now pay one lazy map build above the scan threshold, then reuse O(1) map probes for repeated membership checks. Small inline objects keep the cheap linear scan path.

References:

• Replaces fix: avoid linear objectHas scans for large inline imports #1061.
• Follow-up to fix: O(N^2) materialize for large imported JSON objects with a super #1060.