🔍 Smarter food search that understands synonyms

Improve the food matching algorithm to handle descriptor synonyms, reject conflicting descriptors, and prioritize exact name matches over partial matches from higher-priority sources.

Problem: Currently "butter unsalted" doesn't match "butter without salt" because the matching algorithm doesn't recognize equivalent food descriptors.

Solution: Implement synonym groups for common food descriptors:

Problem: The synonym matching normalizes "unsalted" → no_salt, but doesn't check for conflicting descriptors. "salted" is the opposite of "unsalted" and should be rejected.

Examples:

• ●Query: "butter unsalted" should NOT match "Butter, salted"
• ●Query: "unsweetened tea" should NOT match "sweetened tea"

Solution: Add conflict detection - if query contains a descriptor, reject foods with the opposite descriptor:

• ●no_salt conflicts with salted
• ●no_sugar conflicts with sweetened
• ●fat_free conflicts with full_fat
• ●raw conflicts with cooked methods (grilled, fried, baked, boiled)

Problem: "200g cream" matches "Cheese, cream" from My Foods instead of "Cream" from USDA dataset.

Current behavior: The priority system (My Foods > Dataset > OFF) always prefers higher-priority sources regardless of match quality.

Expected behavior: Exact or near-exact name matches in lower-priority sources should beat partial matches in higher-priority sources.

Example: When searching "cream", "Cream" from USDA should rank higher than "Cheese, cream" from My Foods because "Cream" is an exact match.

Problem: Searching "milk skim" matches "Yoghurt, plain" with brand "skim milk" instead of actual skim milk products.

Solution:

1. ●Deprioritize brand matches in isStrongMatch() - brand should only be a tiebreaker, not a primary match criterion
2. ●In relevanceScore(), give lower scores to brand-only matches
3. ●Consider excluding brand from single-word query matching entirely

1. ●Add descriptorSynonymGroups dictionary in FoodMatchingUtils.swift
   • ●Create a data structure mapping each synonym to its canonical form
   • ●Example: ["unsalted": "no_salt", "without salt": "no_salt", "no salt": "no_salt", "salt free": "no_salt"]
2. ●Add descriptorConflicts dictionary to define mutually exclusive pairs
   • ●Example: ["no_salt": ["salted"], "salted": ["no_salt"]]
3. ●Create normalizeDescriptor() function to map synonyms to canonical forms
   • ●Should handle compound descriptors (e.g., "unsalted low fat")
   • ●Case-insensitive matching
4. ●Create haveMatchingDescriptors() function to detect conflicts
   • ●Returns false if query and food have conflicting descriptors
5. ●Update isStrongMatch() and relevanceScore() to use synonym matching and conflict detection
   • ●Normalize both query and candidate descriptors before comparison
   • ●Treat synonym-matched descriptors as equivalent for scoring purposes
   • ●Reject matches with conflicting descriptors
6. ●Modify pickBestCandidate() to consider match quality across all sources before applying priority
   • ●Introduce a match quality threshold (e.g., exact match = 1.0, strong match = 0.9, partial = 0.5)
   • ●Only apply source priority as a tiebreaker when match quality is similar
   • ●Proposed logic: finalScore = matchQuality * 100 + sourcePriority (ensures quality dominates)

• ●NutriKit/Voice/FoodMatchingUtils.swift - Main implementation location
• ●NutriKit/Voice/TextLoggingModel.swift - May need updates if search flow changes affect the model layer

• ● Synonym matching: "butter unsalted" matches "butter without salt" with high confidence
• ● Synonym matching: "unsweetened yogurt" matches "yogurt no sugar"
• ● Conflict detection: "butter unsalted" does NOT match "Butter, salted"
• ● Conflict detection: "unsweetened tea" does NOT match "sweetened tea"
• ● Exact match priority: Searching "cream" returns "Cream" from USDA over "Cheese, cream" from My Foods
• ● Exact match priority: Searching "chicken" returns "Chicken" over "Chicken salad" when exact match exists
• ● Brand deprioritization: "milk skim" matches milk products, not yogurt with "skim milk" brand
• ● Brand deprioritization: Name/detail matches always rank higher than brand-only matches
• ● Existing exact matches continue to work correctly (no regressions)
• ● Unit tests added for synonym normalization and conflict detection logic

Follow-up to PXL-826 (Voice/Text Logging food matching fix) Incorporates requirements from PXL-837 (now canceled as duplicate)

Build instruction: Use -destination 'platform=iOS Simulator,name=iPhone 17 Pro' when building this project.