feat: New TableFormer model weights

[cau-git]

• New Tableformer weights that work better for tables that have lots of text
• This PR sets "Accurate" tableformer mode as default mode
• Re-generated tests
• Evaluations show improvement on every evaluation dataset!

Checklist:

• Documentation has been updated, if necessary.
• Examples have been added, if necessary.
• Tests have been added, if necessary.

---

Evaluation of FinTabNet

FinTabNet table_structure TEDS_struct-with-text:
mean=0.90 median=0.92 std=0.09
vs mean=0.89 median=0.91 std=0.10

FinTabNet table_structure TEDS_struct-only:
mean=0.90 median=0.92 std=0.08
vs mean=0.90 median=0.92 std=0.09

---

Evaluation of PubTabNet

PubTabNet table_structure TEDS_struct-with-text:
mean=0.85 median=0.90 std=0.17
vs mean=0.80 median=0.83 std=0.13

PubTabNet table_structure TEDS_struct-only:
mean=0.86 median=0.90 std=0.15
vs mean=0.81 median=0.84 std=0.12

---

Evaluation of Pub1M

Pub1M table_structure TEDS_struct-with-text:
mean=0.82 median=0.87 std=0.18
vs mean=0.82 median=0.86 std=0.17

Pub1M table_structure TEDS_struct-only:
mean=0.84 median=0.88 std=0.17
vs mean=0.84 median=0.87 std=0.16

[mergify]

Merge Protections

Your pull request matches the following merge protections and will not be merged until they are valid.

🟢 Enforce conventional commit

Wonderful, this rule succeeded.

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

🟢 Require two reviewer for test updates

Wonderful, this rule succeeded.

When test data is updated, we require two reviewers

• #approved-reviews-by >= 2

[maxmnemonic]

Tests don't break with new TF weights, so no need to update them!
(Edit: it's because tests run FAST configuration, not ACCURATE, I'm changing that...)

[maxmnemonic]

Switched TF model to Accurate by default (in API and CLI), and re-generated tests

[dolfim-ibm]

@maxmnemonic after release, let's push the performance details in the docs.