fix: move punctuation logic to LLM - backend returns complete joined text

The frontend no longer does separator/punctuation logic. Both Maestro
modes (generate and complete) now simply setPrompt(result). The LLM
is instructed to return the user's exact words + continuation, properly
joined. This is the right layer for text joining decisions.

backend/src/text_gen.py

@@ -127,18 +127,15 @@ Output ONLY the prompt text. No numbers, no labels, no commentary."""
 
 MAESTRO_COMPLETE_PROMPT = """You are Maestro, an expert music prompt writer for AI music generation models (specifically Google Lyria).
 
-The user has written the beginning of a prompt and wants you to continue it. Here is what they have:
+The user has started writing a prompt and wants you to continue it. Here is what they have:
 
 "{user_input}"
 
-Continue and expand their idea. Add 2-3 sentences of specific musical detail — genre, instrumentation, mood, tempo, structure, production style. Your continuation will be appended directly after their text, so:
+Continue their idea by adding 2-3 sentences of specific musical detail — genre, instrumentation, mood, tempo, structure, production style.
 
-- Do NOT repeat or rephrase what they wrote
-- Do NOT add a period if their text ends mid-sentence
-- Start naturally as if picking up from where they left off
-- Be vivid and specific — a producer should be able to generate from this
+Return the COMPLETE prompt: the user's EXACT words, followed by your continuation, properly joined. Do not rephrase, rewrite, or change any of the user's words. Just continue from where they left off.
 
-Output ONLY your continuation text. No labels, no commentary, no quotes."""
+Output ONLY the complete prompt text. No labels, no commentary, no quotes."""
 
 
 MAESTRO_STRUCTURED_PROMPT = """You are Maestro, an expert music prompt writer for AI music generation models (specifically Google Lyria).