fix(utf8): prevent multi-byte utf8 characters from being mangled (#981)

**Description**

This PR fixes #677 using [suggested
solution](https://github.com/go-skynet/LocalAI/issues/677#issuecomment-1695939097)
from @yantoz

before:
```
❯ curl -N http://localhost:57541/v1/completions -H "Content-Type: application/json" -d '{
     "model": "ggml-model-q4_0.bin",
     "prompt": "",
     "max_tokens": 32,
     "temperature": 0.7,
     "stream": true
   }'
data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"\ufffd"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"\ufffd"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"\ufffd"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"\ufffd"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":" |"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":" I"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"'"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"text":"m"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
```

now:
```
❯ curl -N http://localhost:57541/v1/completions -H Content-Type: application/json -d {
   "model": "ggml-model-q4_0.bin",
   "prompt": "",
   "max_tokens": 32,
   "temperature": 0.7,
   "stream": true
 }
data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":"😂"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":" "}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":"|"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":" "}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":"I"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":"'"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

data: {"object":"text_completion","model":"ggml-model-q4_0.bin","choices":[{"index":0,"text":"m"}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
```

**Notes for Reviewers**


**[Signed
commits](../CONTRIBUTING.md#signing-off-on-commits-developer-certificate-of-origin)**
- [X] Yes, I signed my commits.
 

<!--
Thank you for contributing to LocalAI! 

Contributing Conventions:

1. Include descriptive PR titles with [<component-name>] prepended.
2. Build and test your changes before submitting a PR. 
3. Sign your commits

By following the community's contribution conventions upfront, the
review process will
be accelerated and your PR merged more quickly.
-->

Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>

api/backend/llm.go

@@ -6,6 +6,7 @@ import (
 	"regexp"
 	"strings"
 	"sync"
+	"unicode/utf8"
 
 	config "github.com/go-skynet/LocalAI/api/config"
 	"github.com/go-skynet/LocalAI/api/options"
@@ -97,9 +98,23 @@ func ModelInference(ctx context.Context, s string, loader *model.ModelLoader, c
 
 		if tokenCallback != nil {
 			ss := ""
-			err := inferenceModel.PredictStream(ctx, opts, func(s []byte) {
+
-				tokenCallback(string(s), tokenUsage)
+			var partialRune []byte
-				ss += string(s)
+			err := inferenceModel.PredictStream(ctx, opts, func(chars []byte) {
+				partialRune = append(partialRune, chars...)
+
+				for len(partialRune) > 0 {
+					r, size := utf8.DecodeRune(partialRune)
+					if r == utf8.RuneError {
+						// incomplete rune, wait for more bytes
+						break
+					}
+
+					tokenCallback(string(r), tokenUsage)
+					ss += string(r)
+
+					partialRune = partialRune[size:]
+				}
 			})
 			return LLMResponse{
 				Response: ss,