refactor(detekt): extract magic-number constants, wrap long lines, drop legit suppressions

Mechanical detekt cleanup across apps/core/infrastructure (behavior-preserving):
- MagicNumber 62->10: named constants + removed 'legit' MagicNumber suppressions
  (Tier level from ordinal, ReplayInferenceProvider CHARS_PER_TOKEN, ConfigLoader
  DEFAULT_* consts, capability scores, HTTP ranges, column widths, etc.)
- MaxLineLength cut to ~0 in production modules (line wraps, no logic change)
- Dropped one dead parameter (ConfigLoader.parseArray lineNum)

Structural findings (ReturnCount/LongMethod/Complexity/LargeClass) left for a
deliberate decomposition pass. Full build + detekt gate green.

core/context/src/main/kotlin/com/correx/core/context/builder/DefaultContextPackBuilder.kt

@@ -97,8 +97,11 @@ class DefaultContextPackBuilder(
         // Stage 1 (FORMAT_COMPRESS): lossless json→dotted-line compaction of structured entries.
         val formatted = if (policy.enabled(CompressionStage.FORMAT_COMPRESS)) {
             deduped.map { entry ->
-                if (classifier.classify(entry) == ContextClass.STRUCTURED) reencode(entry, formatCompressor.compress(entry.content))
-                else entry
+                if (classifier.classify(entry) == ContextClass.STRUCTURED) {
+                    reencode(entry, formatCompressor.compress(entry.content))
+                } else {
+                    entry
+                }
             }
         } else deduped
 

core/context/src/test/kotlin/CompressionPipelineStagesTest.kt

@@ -142,7 +142,11 @@ class CompressionPipelineStagesTest {
         // Instruction-doc pruning is retired: curated/procedural docs (project profile, AGENTS.md
         // how-to) must reach the model verbatim — token-pruning fused them into unparseable soup.
         val pruner = object : com.correx.core.context.compression.TokenPruner {
-            override suspend fun prune(content: String, protectedSpans: List<String>, targetRatio: Double): String = "DOC-PRUNED"
+            override suspend fun prune(
+                content: String,
+                protectedSpans: List<String>,
+                targetRatio: Double,
+            ): String = "DOC-PRUNED"
         }
         val builder = com.correx.core.context.builder.DefaultContextPackBuilder(
             com.correx.core.context.compression.DefaultContextCompressor(), CompressionPolicy(3), tokenPruner = pruner,
@@ -155,8 +159,11 @@ class CompressionPipelineStagesTest {
             entry("agi", ContextLayer.L0, EntryRole.SYSTEM, bigInstructions).copy(sourceType = "agentInstructions"),
         )
         val flat = builder.build(
-            com.correx.core.events.types.ContextPackId("p"), com.correx.core.events.types.SessionId("s"),
-            com.correx.core.events.types.StageId("st"), entries, com.correx.core.context.model.TokenBudget(limit = 4000),
+            com.correx.core.events.types.ContextPackId("p"),
+            com.correx.core.events.types.SessionId("s"),
+            com.correx.core.events.types.StageId("st"),
+            entries,
+            com.correx.core.context.model.TokenBudget(limit = 4000),
         ).layers.values.flatten()
         assertEquals(bigProfile, flat.first { it.sourceType == "projectProfile" }.content)
         assertEquals("you are an agent", flat.first { it.sourceType == "systemPrompt" }.content)
@@ -171,7 +178,10 @@ class CompressionPipelineStagesTest {
         assertEquals(ContextClass.STRUCTURED, c.classify(entry("toolLog", ContextLayer.L2, EntryRole.TOOL)))
         assertEquals(ContextClass.STRUCTURED, c.classify(entry("steeringNote", ContextLayer.L2, EntryRole.SYSTEM)))
         // Assistant tool-call turns are structured history, never token-pruned (amnesia-loop fix).
-        assertEquals(ContextClass.STRUCTURED, c.classify(entry("assistantToolCall", ContextLayer.L2, EntryRole.ASSISTANT)))
+        assertEquals(
+            ContextClass.STRUCTURED,
+            c.classify(entry("assistantToolCall", ContextLayer.L2, EntryRole.ASSISTANT)),
+        )
         assertEquals(ContextClass.FREEFORM, c.classify(entry("chat", ContextLayer.L1, EntryRole.USER)))
     }
 }