Chore(doc): merge multitask training doc (deepmodeling#4395)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **Documentation**
- Updated multi-task mode documentation to clarify the correct link for
model freezing.
- Enhanced fine-tuning documentation for TensorFlow and PyTorch, adding
clarity on processes and configurations.
- Consolidated multi-task training references in the documentation for
easier navigation.
- Removed deprecated TensorFlow multi-task training documentation,
redirecting users to the PyTorch backend.
- Revised multi-task training documentation to emphasize the transition
to PyTorch as the sole supported backend.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

doc/freeze/freeze.md

@@ -24,7 +24,7 @@ $ dp --pt freeze -o model.pth
 
 in the folder where the model is trained. The output model is called `model.pth`.
 
-In [multi-task mode](../train/multi-task-training-pt.md), you need to choose one available heads (e.g. `CHOSEN_BRANCH`) by `--head`
+In [multi-task mode](../train/multi-task-training), you need to choose one available heads (e.g. `CHOSEN_BRANCH`) by `--head`
 to specify which model branch you want to freeze:
 
 ```bash

doc/train/finetuning.md

@@ -129,7 +129,7 @@ Then, prepare a suitable input script for multitask fine-tuning `multi_input.jso
 
 - Suppose the new dataset for fine-tuning is named `DOWNSTREAM_DATA`, and the datasets to be retained from multitask pre-trained model are `PRE_DATA1` and `PRE_DATA2`. One can:
 
-1. Refer to the [`multi-task-training`](./multi-task-training-pt.md) document to prepare a multitask training script for two systems,
+1. Refer to the [`multi-task-training`](./multi-task-training) document to prepare a multitask training script for two systems,
    ideally extracting parts (i.e. {ref}`model_dict <model/model_dict>`, {ref}`loss_dict <loss_dict>`, {ref}`data_dict <training/data_dict>` and {ref}`model_prob <training/model_prob>` parts) corresponding to `PRE_DATA1` and `PRE_DATA2` directly from the training script of the pre-trained model.
 2. For `DOWNSTREAM_DATA`, select a desired branch to fine-tune from (e.g., `PRE_DATA1`), copy the configurations of `PRE_DATA1` as the configuration for `DOWNSTREAM_DATA` and insert the corresponding data path into the {ref}`data_dict <training/data_dict>`,
    thereby generating a three-system multitask training script.

doc/train/index.rst

@@ -8,8 +8,7 @@ Training
    training-advanced
    train-input
    parallel-training
-   multi-task-training-tf
-   multi-task-training-pt
+   multi-task-training
    tensorboard
    gpu-limitations
    finetuning

doc/train/multi-task-training-pt.md → doc/train/multi-task-training.md

@@ -4,7 +4,9 @@
 **Supported backends**: PyTorch {{ pytorch_icon }}
 :::
 
-<!-- we plan to drop TensorFlow backend multi-task training. Replace with the PyTorch one -->
+:::{warning}
+We have deprecated TensorFlow backend multi-task training, please use the PyTorch one.
+:::
 
 ## Theory