[codex] Fix download remaining and speed metrics

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD

[IchenDEV]

Summary

• Estimate LLM and ASR download totals from model catalog hints instead of treating Hub progress units as bytes.
• Measure completed bytes from the local model repository directory during downloads.
• Hide sub-byte speed noise so the UI no longer shows misleading 0 B/s or tiny remaining values like 2 B.
• Add focused tests for estimate parsing and speed formatting.

Root Cause

The model download callback reports logical progress units for some Hub/MLX downloads, not byte counts. The UI was using those unit counts as bytes, which made remaining size and speed incorrect near the end of a download.

Validation

• swift test
• git diff --check origin/main...HEAD