Add Bengali exercise instructions

[rarhs]

Adds Bengali (bn) as the 10th instruction language, following the Hindi (#42) and Polish/Korean (118e4bd) convention: nested instructions.bn + instruction_steps.bn on all 1,324 records, no flat *_bn fields.

Updates data/exercises.json, the embedded EXERCISES blob and SQL schemas/INSERT builder in index.html, the same in setup.html plus its LLM-prompt field list, and the README (9 → 10 languages).

Verification: validates against data/exercises.schema.json (1,324 records, zero errors); both copies of the data are byte-identical; purely additive — stripping bn reproduces the previous data byte-for-byte, with every attribution field intact.

Translations are AI-assisted and haven't had a native-speaker review yet — happy to revise terminology or register if you have preferences.

[riju568]

1. JSON Data & Schema Validation Failures
   The main components of the project lie in data/exercises.json, which should validate against the schema in data/exercises.schema.json, and the common mistakes include:

(a.)Exercise Language Not Found:If you added or altered an exercise instruction at instruction., then you have also toadd an identical instruction string arrayatinstruction_steps.. If the one exist and the other don't then the schema validator and the build will break.

(b.)Schema Drift: If you are introducing a completely new language (like French fr), you have to update data/exercises.schema.json to allow fr under the language pattern maps. If you don't, the automated dataset validator (added in PR #36) will reject it.

(c.)Trailing Commas / Invalid JSON Syntax. One missing or misplaced comma in a large 1,324 element JSON file would prevent it from being parsed entirely.

2. Media Asset & Path Issues:
   This dataset uses very strict matching rules for both the animation GIF and thumbnail media:

(a.)Path Format: The value for the image and gifurl fields within the JSON structure need to be represented as relative paths. (E.g., images/0001-2gPfomN.jpg rather than http://site.net/images/0001-2gPfomN.jpg) If you fail to align the image name along with the mediaid field, you'll get failed image rendering issues.

3. Front-end Integration (index.html & setup.html)
   If the PR touches the interactive browser or developer setup tools:

(a.)UI Dropdown Sync: Whenever a language is added, you need to update the menus inside index.html for language selection and filtering by client side

(b.)Template Mismatch: The client-side script that auto-generates SQL for users will break if either the structure of the database generation script or the live LLM prompt in setup.html is modified.

[rarhs]

Thanks for the checklist, @riju568! Quick summary of where this PR stands on each point:

1. Data & schema — All 1,324 records carry both instructions.bn and instruction_steps.bn, in both copies of the dataset (data/exercises.json and the EXERCISES blob embedded in index.html), and I've verified the two copies are identical. The schema now explicitly lists bn as a required language, following the precedent set by Add hi, pl, ko to JSON Schema language maps #43 and Add French (fr) exercise instructions #44. One small correction, though: the schema's language maps accept new languages via additionalProperties (that's how hi/pl/ko validated before Add hi, pl, ko to JSON Schema language maps #43 listed them), and the validator from Add dataset validator #36 hasn't been merged, so there's no automated check in CI here yet — I validated locally against data/exercises.schema.json with jsonschema: 0 errors.

2. Media — This PR is instructions-only; no media files or image/gif_url/media_id fields are touched.

3. Front-end — bn is registered in index.html (LANG_LABELS + the langs array, which drive the language tabs in the detail modal), and setup.html now includes instructions_bn in all four SQL dialect templates, the INSERT generator, and the LLM prompt. While doing that I noticed Add French (fr) exercise instructions #44 had added French to the LLM prompt but not to the SQL generation, so the generated SQL was silently dropping French text — this PR adds the missing instructions_fr columns alongside bn to keep the tool consistent.

Also worth noting: after #44 merged, this branch conflicted with main (French touched every record), so I've rebuilt it on top of the current main — it's conflict-free and mergeable again as of the latest push.