[Feature] Optimize partition discovery with projected manifest scans

### Search before asking

- [x] I searched in the [issues](https://github.com/apache/paimon/issues) and found nothing similar.

### Motivation

Partition discovery only requires a small subset of manifest fields, but `readPartitionEntries()` currently reads complete manifest entries. This causes unnecessary decoding and object allocation for fields such as key/value statistics and embedded indexes.

Although partition entries are created eagerly, the scan still builds an intermediate list for each manifest before merging the partition statistics.

### Solution

Introduce a projected, streaming manifest scan for partition discovery:

- Project only the fields required for partition aggregation and structural filtering.
- Push partition and bucket filters into the manifest reader.
- Merge projected entries directly into the partition map without per-manifest intermediate lists.
- Fall back to complete manifest entries when active filters require additional fields.
- Preserve the existing cache path for cache-eligible manifests.

This change does not modify the manifest format or partition scan semantics.

### Anything else?

_No response_

### Are you willing to submit a PR?

- [x] I'm willing to submit a PR!

Start by tracing readPartitionEntries() into the manifest reader and identify the fields needed for partition aggregation and structural filtering. Verify how partition and bucket filters, cache-eligible manifests, and the partition map are handled; done means projected entries stream directly into the map while filter-dependent cases still use complete entries and existing scan semantics remain unchanged.

Written by the indexing model from the issue text.