refactor: migrate to URL.parse()

[stevenjoezhang]

check list

• Add test cases for the changes.
• Passed the CI test.

Description

Migrate the remaining legacy url.parse() call sites to the non-throwing WHATWG URL.parse() API, now that hexo-util requires Node.js 22.1 or newer after #449.

• Parse complete absolute URL inputs in encodeURL() and decodeURL(), while preserving the fast path for ordinary relative paths.
• Replace new URL() exception handling with explicit null handling.
• Use WHATWG hostname parsing in isExternalLink(), full_url_for(), and url_for().
• Treat malformed URLs as fallback input instead of throwing, and support IPv6 site hosts correctly.

Issue resolved: hexojs/hexo#3846.

Additional information

Local checks:

• TypeScript build and 266 tests on Node.js 22.1.0
• TypeScript build and 266 tests on Node.js 24.15.0
• TypeScript build and 266 tests on Node.js 26.7.0
• npm run eslint
• 100% line coverage for all five modified runtime files

[SukkaW]

BTW how benchmark shows?

[stevenjoezhang]

@SukkaW I benchmarked hexo-util before and after this PR, with Node.js v26.7.0 and Apple M1 Pro, 32 GiB RAM. The results:

Workload	Before	After	Change
encodeURL() — absolute URL	224,474 ops/s	235,755 ops/s	+5.0%
encodeURL() — relative-path fast path	1,796,595 ops/s	1,803,873 ops/s	+0.4%
decodeURL() — absolute URL	407,214 ops/s	444,515 ops/s	+9.2%
decodeURL() — relative-path fast path	2,569,560 ops/s	2,558,090 ops/s	−0.4%
isExternalLink() — cache miss	541,076 ops/s	674,181 ops/s	+24.6%
full_url_for() — cache miss	170,467 ops/s	190,514 ops/s	+11.8%
url_for() — cache miss	312,992 ops/s	365,563 ops/s	+16.8%

The ±0.4% differences in the relative-path fast paths are within normal run-to-run variance, so those paths can be considered unchanged.