perf: replace i18n pointer tables with offset tables, strip unused strings (#1408)

## Summary

* **What is the goal of this PR?** Reduce the flash footprint of the
i18n string data and tooling improvements to `gen_i18n.py`.

* **What changes are included?**

### 1. `lib/I18n/I18n.cpp` — use offset-based lookup

`I18n::get()` previously dereferenced a `const char* const*` pointer
array. It now uses a two-field `LangStrings` struct (a flat char blob +
a `uint16_t` offset table) generated for each language:

```cpp
// before
const char* const* strings = getStringArray(_language);
return strings[index];

// after
const LangStrings lang = getLanguageStrings(_language);
return lang.data + lang.offsets[index];
```

Lookup cost is unchanged — still O(1), one array load and one addition.

### 2. `scripts/gen_i18n.py` — new generated layout

Each language's string data is now emitted as:

- **`STRINGS_XX_DATA[]`** — a single `const char[]` blob of all strings
concatenated with `\0` separators.
- **`OFFSETS_XX[]`** — a `uint16_t` array of one byte-offset per `StrId`
into the blob.

Previously each language had a `const char* const STRINGS_XX[]` pointer
array (4 bytes/entry on ESP32-C3).

#### Flash savings

| Table type | Size per language | 19 languages |
|---|---|---|
| `const char*` pointer array (before) | `339 × 4 = 1,356 B` | **25,764
B** |
| `uint16_t` offset table (after) | `339 × 2 = 678 B` | **12,882 B** |
| **Saved** | | **12,882 B (~12.6 KB)** |

String data size is unchanged — 133,092 B across 19 languages.

**Total: 158,856 B → 145,974 B** (pointer tables → offset tables).

### 3. Build-time stripping of unused strings

`gen_i18n.py` now scans the `src/` and `lib/` trees for `STR_*`
references and, during a PlatformIO build, automatically omits the 52
strings that are defined in YAML but never referenced in code. This
further reduces the compiled output from 339 → 287 string keys per
language.

---

## `gen_i18n.py` CLI reference

```
python gen_i18n.py [translations_dir [output_dir]] [options]
```

| Argument / Flag | Default | Description |
|---|---|---|
| `translations_dir` | `lib/I18n/translations` | Path to the
per-language YAML files |
| `output_dir` | `lib/I18n/` | Where to write the generated `.h` /
`.cpp` files |
| `--src-dirs DIR [DIR …]` | `src lib` | Directories scanned for `STR_*`
usage |
| `--strip-unused` | off | Remove unreferenced `STR_*` keys from
generated output |
| `--verbose` / `-v` | off | Print per-key INFO/WARNING messages and
`Generated:` lines |

The PlatformIO build (SCons `pre:` hook) calls `main(strip_unused=True)`
automatically, so unused strings are always stripped from firmware
builds without any manual flag.

**Default run output** (no flags) shows a per-language summary table:
```
Language            Code  Own  Fallback  Unused  Data (B)
------------------  ----  ---  --------  ------  --------
English             EN    339  0         52        5,266
Belarusian          BE    310  29        52        9,673
…

  Total: 339  |  Used in code: 287  |  Never used: 52
  Flash (now):    133,092 B strings  +  12,882 B offset tables (uint16_t)  =  145,974 B
  Flash (before): 133,092 B strings  +  25,764 B pointer tables (ptr32)    =  158,856 B
  Saved by offset tables: 12,882 B
```

---

### AI Usage

Did you use AI tools to help write this code? **YES** (GitHub Copilot)

---------

Co-authored-by: Zach Nelson <zach@zdnelson.com>

lib/I18n/I18n.h

@@ -22,6 +22,7 @@ class I18n {
 
   Language getLanguage() const { return _language; }
   void setLanguage(Language lang);
+  const char* getLanguageCode(Language lang) const;
   const char* getLanguageName(Language lang) const;
 
   void saveSettings();