Fix progress bar display to correctly handle iterable dataset and max_steps during training

[bandpooja]

What does this PR do?

This PR fixes the progress bar display in PyTorch Lightning to correctly handle the case when max_steps is set and max_epochs is -1 (infinite epochs mode). Previously, the progress bar did not accurately reflect the total number of batches to process when training was limited by max_steps, causing confusing or incomplete progress updates.

Fixes #20862 and #20124

Does this PR introduce any breaking changes?

No breaking changes introduced. This is a UI/progress bar improvement only.

---

Additional notes

LOCAL TESTS

1. max_steps > total training batches

import torch
from torch.utils.data import Dataset, DataLoader
import torch.nn.functional as F
import pytorch_lightning as pl
from torch import nn
from pytorch_lightning import Trainer

# Dummy Dataset
class DummyDataset(Dataset):
    def __len__(self):
        return 10000  # Large enough to allow many steps

    def __getitem__(self, idx):
        x = torch.randn(10)
        y = torch.randint(0, 2, (1,))
        return x, y[0]

# Simple Model
class DummyModel(pl.LightningModule):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 2)

    def forward(self, x):
        return self.linear(x)

    def training_step(self, batch, batch_idx):
        x, y = batch
        logits = self(x)
        loss = F.cross_entropy(logits, y)
        self.log("train_loss", loss)
        return loss

    def configure_optimizers(self):
        return torch.optim.Adam(self.parameters(), lr=1e-3)

dataset = DummyDataset()
loader = DataLoader(dataset, batch_size=32)

model = DummyModel()

trainer = Trainer(
    max_steps=500,              # 👈 This tests the changes!
    accelerator="cpu",
    log_every_n_steps=1,
    enable_model_summary=False,
)

trainer.fit(model, train_dataloaders=loader)

2. max_steps < total training batches

trainer = Trainer(
    max_steps=100,              # 👈 This tests the changes!
    accelerator="cpu",
    log_every_n_steps=1,
    enable_model_summary=False,
)

trainer.fit(model, train_dataloaders=loader)

3. Training with iterable dataset as done in Why does the progress bar not show the total steps when using iterable dataset? #20124

from torch.utils.data import IterableDataset

class InfiniteIterableDataset(IterableDataset):
    def __iter__(self):
        while True:
            x = torch.randn(10)
            y = torch.randint(0, 2, (1,))
            yield x, y[0] # infinite stream

dataset = InfiniteIterableDataset()
loader = DataLoader(dataset, batch_size=32)

model = DummyModel()

trainer = Trainer(
    max_steps=500,              # 👈 This tests the changes!
    accelerator="cpu",
    log_every_n_steps=1,
    enable_model_summary=False,
)

trainer.fit(model, train_dataloaders=loader)

[Borda]

could we have a test that this return the correct number, pls

[bandpooja]

@Borda, I'm still getting up to speed with writing effective tests — happy to hear any feedback!

[Borda]

let's have assert on progress_bar property: total_train_batches

[Borda]

I'm still getting up to speed with writing effective tests — happy to hear any feedback!

overall the test looks good, let's just have a direct assert on the property