[lightning] Lightning importinto can skip new input because of a finished checkpoint from an older run

## Bug Report

> Evidence status: confirmed.

### 1. Minimal reproduce step (Required)

Retain a finished checkpoint for db.t with the supported keep-after-success strategy or reuse its checkpoint DSN; start another run for db.t with a different nonempty input file; observe SubmitAndWait result and SubmitTable count.

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

The checkpoint is accepted only when its input/config/target fingerprint matches the current run; otherwise the table is submitted or the run fails closed.

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

SubmitAndWait returns nil and logs no jobs to execute while SubmitTable is called zero times for the new nonempty input.

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

master 13282a8bd06b

Likely root cause and fix direction

TableCheckpoint is keyed by table name and stores status/job/group but no input fingerprint; the Finished fast path skips unconditionally. The importinto success path implements only checkpoint removal and leaves rename/origin checkpoints reusable in place.

Start with TableCheckpoint and the importinto success path described in the report; trace the Finished fast path used by SubmitAndWait and the checkpoint removal or rename handling. Reproduce with a retained checkpoint for db.t and a different nonempty input, then verify the new run is submitted or fails closed and SubmitTable is no longer skipped.

Written by the indexing model from the issue text.