Add feature to automatically convert external images to S3 (#41)

[l3ickey]

This PR adds a new feature to automatically upload external image links in Obsidian notes to S3 and replace them.

Key changes:

• Add a new command "Convert External Images to S3"
• Execute the process in 3 stages:
  　1. Replace ALT text in external image links with file hash
  　2. Run Obsidian's "Download attachments for current file" command
  　3. Upload downloaded images to S3 and replace original links
• Implement automatic detection of download completion
• Automatically remove duplicate S3 links

Example of the transformation process:

Initial external image: ![image](https://example.com/img.png)

1. After Step 1 (hash generation): ![a82b844446ae6b9bb95c47d806265e4a.png](https://example.com/img.png)
2. After Step 2 (download): Both ![[img.png|a82b844446ae6b9bb95c47d806265e4a.png]] and ![image](https://obsidian-bucket.s3.ap-northeast-1.amazonaws.com/a82b844446ae6b9bb95c47d806265e4a.png) exist in the document
3. After Step 3 (cleanup): Only ![image](https://obsidian-bucket.s3.ap-northeast-1.amazonaws.com/a82b844446ae6b9bb95c47d806265e4a.png) remains

This feature makes it easy to migrate notes containing image links from other sources to S3 hosting.

[jvsteiner]

It would be a lot easier for me to review this if you could squash these into a single commit.

[jvsteiner]

I have some concerns:

Architecture:

• Relies on Obsidian's internal (this.app as any).commands.executeCommandById — undocumented API
  that could break
• Polling loop (30s timeout, 1s intervals) to detect download completion is fragile; there's no
  event-based way to know when Obsidian finishes downloading

Code quality:

• Significant code duplication — the editor-finding logic (search leaves, fall back to active view)
  is repeated in processDownloadedImages, replaceAltWithHash, and could be extracted to a helper
• The s3LinkRegex is named "S3" but matches any http(s) URL — misleading naming
• error.message used without type narrowing (TypeScript strict mode issue)
• Several console.log calls that look like debug output left in

Robustness:

• editor.setValue() replaces the entire document, which can lose undo history and is generally
  discouraged in Obsidian plugins — editor.replaceRange() is preferred
• The hash-matching logic assumes filenames are formatted as hash.extension, which could collide
  with unrelated content

Minor:

• Whitespace-only changes to onunload() { } and the ternary formatting are noise — my earlier
  comment about squashing commits applies here