logseq#12539: reduces memory allocation for graph file query to avoid crash; adds debug output

[vr-one]

Fix OOM crash in get_all_files_meta + improve error messages and logging

Problem

Logseq Desktop would crash with EXC_BREAKPOINT (SIGTRAP) on the tokio-runtime-worker thread during file sync on large graphs. The stack trace pointed to an allocation failure inside get_file_meta at graph.rs.

Root cause: get_all_files_meta collects one get_file_meta future per file and passes them all to future::join_all, which polls every future concurrently. Each future immediately allocated a Vec::with_capacity(1024 * 1024) (1 MB) read buffer and held it live for the duration of the I/O loop. With hundreds of files this meant hundreds of MB were reserved simultaneously, exhausting available memory and triggering a Rust OOM panic — which on ARM64 manifests as EXC_BREAKPOINT.

Changes

graph.rs — core fix + diagnostics

• Reduced the read buffer in get_file_meta from 1 MB to 64 KB. The buffer is reset to len = 0 after each chunk, so only capacity is relevant to memory use; 64 KB is more than sufficient for efficient I/O and cuts peak concurrent allocation by 16×.
• Replaced silent unwrap() calls in is_page_file_path with unwrap_or_else(|| panic!(...)) that include the offending path, making future panics immediately actionable.
• Added log::trace! at the entry of get_file_meta to aid per-file debugging.

lib.rs — observability

• get_local_all_files_meta: added a wall-clock timer; logs file count and elapsed milliseconds at INFO level after each full scan (e.g. "get file meta of "/path/to/graph": 347 files in 412ms").
• get_local_files_meta: added log::trace! for the file list being requested.
• rename_local_file: added log::info! showing source and destination paths.

sync.rs — panic quality

• Replaced five bare unwrap() calls with .expect(…) or unwrap_or_else(|| panic!(…)) carrying descriptive context:
  • Proxy::https(proxy) — identifies invalid HTTPS_PROXY URL
  • builder.build() — identifies HTTP client construction failure
  • strip_prefix(bucket()) — shows the mismatched prefix and bucket values
  • self.credentials.as_ref() in upload_tempfile — clarifies that refresh_temp_credential was not called
  • self.s3_prefix.clone() — same

Testing

Verified on macOS arm64 with a ~350-file graph that previously triggered the crash. The app now completes the initial file-sync scan without crashing. The INFO log line confirms the correct file count and a reasonable scan time.

[vr-one]

@andelf @tiensonqin @RCmerci Please review this PR to build a new version of the raspi library and release a maintenance version of logseq - cause the db-version still lacks features to switch for most users.