…tion (#3412)

Inspired by the walrus issue in #3353

## Summary

Several `if x := dict.get(k):` checks treated legitimate falsy values as
missing:

- `lower_bounds.get(field_id)` / `upper_bounds.get(field_id)` return
`bytes`. `b""` is a valid serialization of an empty string and was being
skipped, causing metrics-based row filtering to return
`ROWS_MIGHT_MATCH` when it should have been `ROWS_CANNOT_MATCH` (and
vice versa for strict eval).
- `accessors[...].get(file.partition)` can return `0` or `""` for an
`IdentityTransform` partition. The walrus dropped it, so projected
partition columns were filled with `null` instead of the actual value.
- `inspect.py` `lower_bound` / `upper_bound` rendering had the same
`b""` issue, showing `None` instead of `""` in `readable_metrics`.

All conditions are switched to explicit `is not None` checks. A small
`_readable_bound` helper deduplicates the inspect rendering.

## Changes

- `pyiceberg/expressions/visitors.py` — `_InclusiveMetricsEvaluator` and
`_StrictMetricsEvaluator` bound lookups (21 sites).
- `pyiceberg/io/pyarrow.py` — `_get_column_projection_values` and
`ArrowProjectionVisitor` missing-field handling.
- `pyiceberg/table/inspect.py` — extract `_readable_bound`, use it in
both the `entries` and `_get_files_from_manifest` rendering paths.

## Tests

- `tests/expressions/test_evaluator.py` — inclusive and strict
evaluators with empty-string bounds, covering lower-only, upper-only,
and both-bounds branches.
- `tests/io/test_pyarrow.py` — parametrized identity-transform
projection with falsy partition values (`0`, `""`, `None`).
- `tests/table/test_inspect.py` — `_readable_bound` helper plus
integration tests via `tbl.inspect.entries()` and `tbl.inspect.files()`
for empty-string and null bounds.

Inspired by the walrus issue in #3353

## Problem

`PartitionField` and `SortField` accept the v3 `source-ids` list (added
in [#1554](#1554)) and map
it to the legacy singular `source-id`. Both validators try to reject an
empty list:

```python
if "source-id" not in data and (source_ids := data["source-ids"]):
    if isinstance(source_ids, list):
        if len(source_ids) == 0:
            raise ValueError("Empty source-ids is not allowed")
        ...
        data["source-id"] = source_ids[0]
```

The walrus uses truthiness, and `[]` is falsy — so the `len(source_ids)
== 0` branch is unreachable. Passing `{"source-ids": []}` silently skips
the mapping, and Pydantic then reports a generic "field required" error
instead of the intended message. A missing `source-ids` key also raises
`KeyError` instead of being handled cleanly.

## Fix

Replace the walrus with an explicit key check in both validators:

```python
if "source-id" not in data and "source-ids" in data:
    source_ids = data["source-ids"]
    ...
```

This makes the empty-list validation reachable and avoids the
`KeyError`.

## Tests

Added regression tests that deserialize `{"source-ids": []}` and assert
`ValueError("Empty source-ids is not allowed")` is raised, for both
`PartitionField` and `SortField`.