Computed constants output misrepresents source locations if non-ASCII characters are present in the source

If the script that jsir_gen is acting upon contains non-ASCII characters such as Cyrillic script (even in comments), it mixes up byte-position vs character-position and outputs wrong segments of the script. The issue appears to be that `JsPosition` when fetching from the JSON returns character position, but then `original_source.substr` on lines 75-77 of `jsir_gen_lib.cc` uses `absl::string_view.substr` which operates on bytes.

For example, if you run `jsir_gen` on the script:

```javascript
// тестовые символы кириллицы

let c = atob("SGVsbG8sIHdvcmxkIQ==");
```
It will output:
```
// Dataflow etc omitted
================================================================
Computed constants:
================================================================
From [39, 67): `риллицы

let c = atob` -> `Hello, world!`
================================================================
```
In between the \` \` it has mistakenly captured the newlines and the end of the comment due to the mixup instead of the actual function call.