feat: add context budget system for LLM overflow (#19)

Adds a budget package that estimates token usage and progressively
trims context to fit within model-specific limits.

Trim order (least important first):
1. Language patterns
2. Repository conventions
3. Full file context
4. Diff (truncated as last resort)

When content is trimmed, a note is appended to the user prompt so
the LLM knows context was reduced.

- New budget package with Fit(), EstimateTokens(), LimitForModel()
- Model limit table (GPT-4.1: 128K, GPT-5: 200K, Claude: 200K)
- Refactored review/prompt.go: BuildSystemBase() and BuildUserMeta()
  extract non-trimmable content; old functions delegate to new ones
- main.go uses budget.Fit() instead of direct prompt assembly
- 7 unit tests covering all trim paths

Closes #19

cmd/review-bot/main.go

@@ -10,6 +10,7 @@ import (
 	"strings"
 	"time"
 
+	"gitea.weiker.me/rodin/review-bot/budget"
 	"gitea.weiker.me/rodin/review-bot/gitea"
 	"gitea.weiker.me/rodin/review-bot/llm"
 	"gitea.weiker.me/rodin/review-bot/review"
@@ -141,15 +142,26 @@ func main() {
 		log.Printf("Loaded patterns from %s (%d bytes)", *patternsRepo, len(patterns))
 	}
 
-	// Step 7: Build prompts
-	systemPrompt := review.BuildSystemPrompt(conventions, patterns)
-	userPrompt := review.BuildUserPrompt(pr.Title, pr.Body, diff, fileContext, ciPassed, ciDetails)
+	// Step 7: Budget-aware prompt assembly
+	sections := budget.Sections{
+		SystemBase:  review.BuildSystemBase(),
+		Patterns:    patterns,
+		Conventions: conventions,
+		FileContext: fileContext,
+		Diff:        diff,
+		UserMeta:    review.BuildUserMeta(pr.Title, pr.Body, ciPassed, ciDetails),
+	}
+	budgetResult := budget.Fit(*llmModel, sections)
+	log.Printf("Token estimate: ~%dK (limit: %dK)", budgetResult.EstTokens/1000, budget.LimitForModel(*llmModel)/1000)
+	if len(budgetResult.Trimmed) > 0 {
+		log.Printf("Context trimmed: %v", budgetResult.Trimmed)
+	}
 
 	// Step 8: Call LLM
 	log.Printf("Sending to LLM (%s)...", *llmModel)
 	messages := []llm.Message{
-		{Role: "system", Content: systemPrompt},
-		{Role: "user", Content: userPrompt},
+		{Role: "system", Content: budgetResult.SystemPrompt},
+		{Role: "user", Content: budgetResult.UserPrompt},
 	}
 
 	response, err := llmClient.Complete(ctx, messages)