Fix two long-standing typo bugs in index.js

[dscho]

Two long-standing bugs found while reviewing the codebase in preparation for the imminent Node.js 24 / ESM migration. Both have been in index.js since acdc390 ("Convert the Action to Javascript", 2021-03-20), neither is caught by ESLint, and neither is exercised by the existing test suite, so they have lived undetected through five years of releases.

core.debug$ typo on the content-pattern skip branch

if (contentPattern && !content.match(contentPattern)) {
  core.debug$(`Feed item skipped because it does not match the content pattern (${title})`)
  continue
}

The other five core.debug(...) calls in run() are spelled correctly; only this one carries the stray $. Effect: any workflow
that uses the content-pattern input crashes with TypeError: core.debug$ is not a function the moment a feed item fails the pattern; core.setFailed translates the rejection into a red workflow step.

parseDurationInMilliseconds() triplet regex

The function declares three regex variables (milliSeconds, seconds, minutes) all bound to the same pattern /(\d+)\s*m/, where /(\d+)\s*ms/, /(\d+)\s*s(?!\w)/, and /(\d+)\s*m(?!s)/ were intended. Effect:

input	actual	expected	status
30ms	1,830,030 ms	30 ms	~61,000× off
30s	0 ms	30,000 ms	totally lost
30m	1,830,030 ms	1,800,000 ms	1.7% over
30h	108,000,000	correct	works
30d	2,592,000,000	correct	works

Only h and d happen to land on correct values, and the existing tests use exactly 48h and 9999d, which is why CI never noticed. A user passing max-age: 30s got every feed item filtered out silently.

What this PR does

Two source commits, each followed by an npm run prepare per repo convention:

1. Fix the regex copy-paste, with negative lookaheads that prevent the unit characters from overlapping
   (s(?!\w), m(?!s)). New regression test respects max-age when expressed in seconds pins the seconds case; verified to fail on the buggy code and pass on the fixed code. Verified outputs: 30ms→30, 30s→30000, 30m→1,800,000, 30h→108,000,000, 30d→2,592,000,000, 1h30m→5,400,000, 1d2h→93,600,000.
2. Fix the core.debug$ typo. New regression test respects content-pattern filter configures content-pattern against a two-entry feed where one entry matches and one does not; verified to fail (with the actual TypeError: core.debug$ is not a function) on the buggy code and pass on the fixed code.

[rimrul]

The s(?!\w) guard prevents seconds from being matched inside
plain English words like "30 minutes" (where the s of "minutes"
would otherwise count).

This example in the ca70045 commit message is an odd choice. minutes wouldn't match because e is neither numeric nor whitespace, not because of the '(?!\w)'

[rimrul]

shouldn't the test include at least one entry on either side of the limit to effectively test that the limit is correctly interpreted? at the moment we just test that '30s' isn't interpreted as a vastly to small time span, but it could in theory still be 30 days or even no limit, for all this test catches.