[PerryLink]

What

ComposableConfig.from_dict() silently keeps only the first entry when the dict describes more than
one config:

ComposableConfig.from_dict({
    "fake_a": {"weight_bits": 2},
    "fake_b": {"weight_bits": 6},
})
# -> FakeAConfig(weight_bits=2)          (the second key is gone)
# expected -> ComposableConfig([FakeAConfig(2), FakeBConfig(6)])

The return value is not even a ComposableConfig, so anything that then treats it as one
(config.compose(...), iteration) fails or ignores the other entries. The single-entry case works, which
is why this stayed hidden.

The change

Iterate all entries, build one sub-config per key, and return a ComposableConfig that holds them —
+3/−4 lines in neural_compressor/common/base_config.py. Single-entry behaviour is unchanged.

Verification

• new unit test test_composable_config_from_dict: fails on pristine main with
  FakeAlgoConfig ... is not an instance of ComposableConfig, passes after;
• a direct repro prints FakeAConfig(weight_bits=2) before and
  ComposableConfig([FakeAConfig(2), FakeBConfig(6)]) after;
• test/common suite: 32 tests OK under both unittest and pytest;
• git apply --check on a pristine main; both files are 100644.

No framework-level end-to-end test is included: the bug is in the framework-agnostic common layer, and
torch/TF are not available offline here.

No CLA in the repository; commits are signed off, as the repo's DCO check requires.