fix(tanstack): serialize custom SegmentKey fields in hashSegment (#284)

[JonasJesus42]

Fixes #284. hashSegment only serialized a fixed set of known fields (device, loggedIn, salesChannel, regionId, flags), so any custom dimension a site added via buildSegment was silently dropped from the cache key — requests that should land in different buckets shared one entry and the worker served the wrong cached response, with no compile or runtime error (a real production incident where store/CEP selection didn't vary the PLP cache). Now hashSegment appends any custom key in stable-sorted order while keeping the known fields' historical byte-for-byte format (so existing cache keys are unchanged), SegmentKey carries an index signature documenting the behavior, and the function is hoisted to module scope and exported. Added a describe("hashSegment") suite covering format regression, custom-field variation, stable ordering, array sorting, no flags mutation, and falsy-value omission. Also documented that the HTTP purge endpoint doesn't reach custom dimensions (a pre-existing limitation, already true for flags/tablet).

🤖 Generated with Claude Code

---

Summary by cubic

Fixes #284 by updating hashSegment to include custom SegmentKey fields in the cache key, preventing cross-cohort cache collisions. Known fields keep their exact token format so existing keys remain unchanged.

• Bug Fixes
  • Serialize custom dimensions as key=value in stable-sorted order; arrays sorted; falsy values omitted.
  • Preserve historical tokens for known fields (device|auth|sc=…|r=…|f=…) and avoid mutating flags.
  • Export hashSegment and add tests for format, ordering, and custom fields.
  • Add an index signature to SegmentKey to document custom-dimension behavior.
  • Document that the HTTP purge endpoint still only covers device × salesChannel × regionId (not flags, tablet, or custom keys).

^{Written for commit bf47550. Summary will update on new commits.}