Fix Swift SpO2 capability token normalization

[bhelm]

Summary

• trim whitespace and newlines around Swift capability tokens before filtering
• drop whitespace-only tokens and remove only the bare spo2 token, matching Kotlin canonicalization
• add mirrored Swift/Kotlin coverage for canonical, whitespace-bearing, duplicate-separator, and SpO2-only stored rows

Root cause

Swift converted comma-split substrings directly to String, so restored rows such as hr, spo2,hrv retained the whitespace-bearing SpO2 token and could also lose legitimate whitespace-prefixed metrics when decoded. Kotlin already trims each token and removes empty tokens. The fix applies that same normalization on the real Swift product helper; Kotlin production is unchanged.

Validation

• RED before fix: the unchanged Swift algorithm failed 7 of 9 whitespace/canonicalization cases
• frozen three-file diff is an exact semantic subset of the independently reviewed product-fix union
• Swift focused: 1 test passed
• Swift affected WhoopStore suite: 7 tests passed
• Swift full WhoopStore suite: 420 tests passed, 0 failures
• Kotlin focused/affected: the mirrored regression test passed
• Kotlin FullDebug: 4,157 tests passed, 0 failures/errors, 6 skipped
• git diff --check: pass
• independent static review: pass, no P0/P1/P2

Fixes bhelm#95.