PLAN REPLAYER LOAD fails with "Unknown database" when the dump was taken with tidb_low_resolution_tso enabled

## Bug Report

### 1. Minimal reproduce step (Required)

On cluster A (the dumping side):

```sql
create table t(a int, b int, index idx_a(a));
set @@tidb_low_resolution_tso = 1;
plan replayer dump explain select * from t where a = 1;
```

Download the zip. Its `variables.toml` contains `tidb_low_resolution_tso = "ON"`.

On cluster B (the loading side), with `mysql --local-infile=1`:

```sql
plan replayer load '/path/to/dump.zip';
```

### 2. What did you expect to see? (Required)

The load succeeds, or at least reports the real reason it cannot create the schema.

### 3. What did you see instead (Required)

```
ERROR 1049 (42000): Unknown database 'test'
```

Root cause: `PLAN REPLAYER LOAD` applies every variable from `variables.toml` to the loading session before it creates the schema. With `tidb_low_resolution_tso` on, TiDB rejects the `create database if not exists` DDL with "can not execute write statement when 'tidb_low_resolution_tso' is set". `createSchemaAndItems` in `pkg/executor/plan_replayer.go` logs that error only at debug level and continues, so the following `use ` fails with "Unknown database", which hides the actual cause.

`tidb_snapshot` and `tidb_read_staleness` in the dump cause the same class of problem: they override the read timestamp of the loading session, either rejecting the DDL or making the freshly created schema invisible to a stale infoschema.

### 4. What is your TiDB version? (Required)

Reproduced on v8.5.7 and current master (e73d535e34).

Start in pkg/executor/plan_replayer.go at createSchemaAndItems and reproduce the dump/load sequence using tidb_low_resolution_tso, tidb_snapshot, or tidb_read_staleness. Trace how variables.toml values are applied before schema creation and how schema-creation errors are logged. Done means PLAN REPLAYER LOAD no longer hides the schema-creation failure behind "Unknown database" and reports the real cause or loads successfully.

Written by the indexing model from the issue text.