[BUG] Result inconsistent when intermediate results executed

Nobody has claimed this yet.

**Describe the bug**
The same DataFrame expression outputs different results. The only difference is that the first piece of code does not execute intermediately while the second executes.

**To Reproduce**
```python
import numpy as np
import pandas as pd
import mars.dataframe as md
import mars.tensor as mt

rs = np.random.RandomState(0)
raw_df = rs.rand(20, 10)
raw_df = pd.DataFrame(
np.where(raw_df > 0.4, raw_df, np.nan), columns=list("ABCDEFGHIJ")
)
raw_df2 = rs.rand(20, 10)
raw_df2 = pd.DataFrame(
np.where(raw_df2 > 0.4, raw_df2, np.nan), columns=list("ACDEGHIJKL")
)

df = md.DataFrame(raw_df, chunk_size=4)
df2 = md.DataFrame(raw_df2, chunk_size=6)

joined = md.merge(
df.index.to_frame(), df2.index.to_frame(), how="outer", left_index=True, right_index=True
)
df, df2 = df.reindex(joined.index), df2.reindex(joined.index)

nna_df = df.notna().astype(np.float_)
nna_df2 = df2.notna().astype(np.float_)

df, df2 = df.fillna(0), df2.fillna(0)

print(df.mul(nna_df2, axis=0).sum(axis=0).execute().fetch())
"""
prints

A 0.0
B 0.0
C 0.0
D 0.0
E 0.0
F 0.0
G 0.0
H 0.0
I 0.0
J 0.0
K 0.0
L 0.0
dtype: float64
"""

df.execute()
nna_df2.execute()
print(df.mul(nna_df2, axis=0).sum(axis=0).execute().fetch())
"""
prints

A 4.392781
B 0.000000
C 5.063856
D 5.276695
E 4.752212
F 0.000000
G 5.451037
H 4.206434
I 6.948446
J 2.303648
K 0.000000
L 0.000000
dtype: float64
"""
```

Start by running the provided Python reproduction, focusing on md.merge, DataFrame.reindex, DataFrame.execute, and fetch. Compare the two execution paths and trace why intermediate execution changes the result. Done means both paths produce the same DataFrame output.

Written by the indexing model from the issue text.