addrkit: strict ordinal matching, LLM-variant guard, externalized prompt

- match: require candidate ordinal to equal the query ordinal in fuzzy search
  (deterministic typo recall 73% -> 80%, no wrong houses)
- resolver: compatibleVariant guard — LLM may only fix street spelling, keeping
  the same words, ordinal and house unit; damerau distance for transpositions
- llm: prompt moved out of code (go:embed default_prompt.txt) with
  llm.system_prompt / llm.system_prompt_file overrides; guided JSON parsing
  tolerant of thinking blocks
- address: recognize ordinals appearing after the street name
- tests: first coverage for the LLM adapter and config prompt loading; README

Measured (sample 200): deterministic 94.7-95.8% exact / 0 wrong across 4 seeds;
LLM+guard typo 90.7-89.8% / 0 wrong. LLM fallback latency ~1.2-1.9s.

internal/eval/eval.go

@@ -8,6 +8,7 @@
 package eval
 
 import (
+	"context"
 	"fmt"
 	"math/rand"
 	"sort"
@@ -55,15 +56,15 @@ type Report struct {
 	Categories    map[string]CategoryStat `json:"categories"`
 }
 
-// Run evaluates all cases against the index.
-func Run(ix *match.Index, cases []Case) Report {
+// Run evaluates all cases against the resolver.
+func Run(resolver *match.Resolver, cases []Case) Report {
 	rep := Report{Categories: map[string]CategoryStat{}}
 	for _, c := range cases {
 		st := rep.Categories[c.Category]
 		st.Total++
 		rep.Total++
 
-		res := ix.Resolve(address.Parse(c.Query))
+		res := resolver.Resolve(context.Background(), c.Query)
 		exp := address.Parse(c.Expect)
 
 		exact := false