Add compute_metadata toggle to dataset upload

[Irozuku]

Summary

Adds a toggle to control whether full EDA metadata (correlations, numeric/categorical/text stats, quality score) is computed when uploading a dataset. Computing metadata on large datasets can take several minutes and use significant memory - users can now skip it and upload faster. The toggle is available in all three upload paths: normal file upload (dataloader config sidebar), save-as-dataset from notebook, and hub import.

When the dataset has many columns or many rows, the toggle auto-shows a recommendation alert. If the user keeps it ON over the threshold and submits, a confirmation dialog asks them to confirm or skip. Datasets uploaded without metadata show a notice in the EDA views.

---

Type of Change

• Backend change
• Frontend change
• CI / Workflow change
• Build / Packaging change
• Bug fix
• Documentation

---

Changes (by file)

Backend

• DashAI/back/job/dataset_job.py: branch on params["compute_metadata"] (default True). When False, calls compute_base_metadata() instead of compute_metadata(). Strips inherited extended metadata when skipping for notebook-copy path. Notebooks with no converters reuse source metadata directly.
• DashAI/back/api/api_v1/endpoints/datasets.py: preview_with_types response now includes previewed_bytes so the frontend can estimate total row count before submit.
• DashAI/back/api/api_v1/endpoints/dataset_source.py: docstring notes compute_metadata is honored via body.params passthrough.
• tests/back/api/test_dataset_job_compute_metadata_flag.py: tests for the flag branch (full, base-only, default-true).
• tests/back/api/test_preview_with_types_previewed_bytes.py: test for previewed_bytes in preview response.

Frontend

• DashAI/front/src/components/notebooks/datasetCreation/DataloaderConfigBar.jsx: toggle rendered as FormSchemaFieldCard + Switch in the right sidebar, consistent with "Use Native Types". Does not push through formValues (no preview re-render on toggle).
• DashAI/front/src/components/notebooks/datasetCreation/UploadDatasetSteps.jsx: holds computeMetadata state, passes to DataloaderConfigBar and ConfigureAndUploadDatasetStep.
• DashAI/front/src/components/notebooks/datasetCreation/ConfigureAndUploadDatasetStep.jsx: gates submit on confirm dialog when over threshold with metadata ON.
• DashAI/front/src/components/notebooks/datasetCreation/SaveDatasetModal.jsx: toggle + confirm dialog in save-as-dataset modal. Buttons are "Cancel" and "Upload".
• DashAI/front/src/components/notebooks/notebook/DatasetPreviewNotebook.jsx: forwards compute_metadata through enqueueDatasetJob; failure callback verifies dataset state before showing error snackbar.
• DashAI/front/src/pages/hub/HubImportPage.jsx + HubImportPanel.jsx: toggle in hub import sidebar; compute_metadata included in import params; stripped from preview fetch deps so toggling doesn't re-fetch.
• DashAI/front/src/components/datasets/ComputeMetadataToggle.jsx: shared toggle + recommendation alert (cols > 50 or est. rows > 100k).
• DashAI/front/src/components/datasets/ComputeMetadataConfirmDialog.jsx: confirm dialog with "Cancel" / "Upload" buttons.
• DashAI/front/src/components/DatasetVisualization.jsx: info alert when general_info missing.
• DashAI/front/src/components/notebooks/dataset/QualityAlerts.jsx: fixed hook-order crash when metadata is absent.
• DashAI/front/src/utils/metadataRecommendation.js + .test.js: threshold helpers + unit tests.
• i18n: en, es, de, pt locale files updated with computeMetadata.* keys.

---

Testing

• Upload a narrow dataset (< 50 cols) - toggle ON by default, no alert.
• Upload a wide CSV (> 50 cols) - recommendation alert appears.
• Keep toggle ON and submit over threshold - confirm dialog appears; "Upload" proceeds; "Cancel" closes without submitting.
• Upload with toggle OFF - splits.json has only column_names, total_rows, nan. EDA views show the metadata-missing notice.
• Save-as-dataset from a notebook with no converters and toggle OFF - resulting dataset has base metadata only; with toggle ON and an existing metadata-rich source, source metadata is reused.
• Hub import with toggle OFF - base metadata only.

---

Notes

• compute_metadata defaults to True - existing clients that don't send the flag keep current behavior.
• The previewed_bytes field in preview responses is the size of the uploaded content, not just the rows sampled. This gives a conservative row estimate (may underestimate for sparse files).
• No on-demand recompute endpoint - users who want full metadata after skipping must re-upload with the toggle ON.