Handle SQLite cache errors gracefully instead of aborting upload

Summary:
- Fix sqlite3.OperationalError: disk I/O error in the file handle cache from crashing the entire upload process
  - PersistentCache.get() now catches sqlite3.DatabaseError and returns None (cache miss) instead of propagating the exception
  - Cache failures are logged as warnings so they remain visible for debugging

  Context

  The upload file handle cache is a SQLite database that stores server-returned handles for already-uploaded images. It is purely an optimization to skip re-uploading — if the cache is
  unavailable, the server's fetch_offset endpoint confirms prior uploads at the cost of one extra HTTP round-trip per image.

  Previously, any SQLite error during cache lookup (disk I/O error, corrupt database, etc.) propagated up through _continue_or_fail which treated it as a fatal error, aborting the upload of all
   remaining images. This was observed on Windows with 4 concurrent upload workers rapidly opening/closing SQLite connections to the same cache file.

Author: caglarpir

mapillary_tools/history.py

@@ -118,12 +118,21 @@ def get(self, key: str) -> str | None:
 
         s = time.perf_counter()
 
-        with store.KeyValueStore(self._file, flag="r") as db:
+        try:
+            db_ctx = store.KeyValueStore(self._file, flag="r")
+        except sqlite3.DatabaseError as ex:
+            LOG.warning(f"Failed to open cache database: {ex}")
+            return None
+
+        with db_ctx as db:
             try:
                 raw_payload: bytes | None = db.get(key)  # data retrieved from db[key]
             except Exception as ex:
                 if self._table_not_found(ex):
                     return None
+                if isinstance(ex, sqlite3.DatabaseError):
+                    LOG.warning(f"Cache read error for {key}: {ex}")
+                    return None
                 raise ex
 
         if raw_payload is None: