fix(reply-matcher): require corroboration for partial role-title matches - #2672

…tcher (career-ops-hq#2671)

checkRoleMatch() treated any role-title word longer than 3 characters as a
sufficient match on its own, with no requirement that a second signal
(company name or sender domain) also corroborate it. A tracker row titled
"Talent Acquisition Specialist" splits into ["Talent","Acquisition",
"Specialist"], and any unrelated email containing just one of those words
anywhere (e.g. in a different recruiter's signature line) satisfied the
match and got misattributed to that application.

Two changes:
- Add checkRoleMatchExact(), which only credits a match on the *entire*
  role title (or its Chinese symbol-stripped form) as a contiguous
  substring. This is specific enough to stand alone. A role that is a
  single generic word (e.g. "Recruiter") is excluded from this path too,
  since a whole-role check degenerates into a bare-word check for
  single-word titles.
- matchCandidates() now only counts a partial (significant-word) role
  match when it is corroborated by checkCompanyMatch or a sender-domain
  match already computed for that application. checkRoleMatch() also
  gains a stoplist (GENERIC_ROLE_WORDS) of generic recruiting/HR
  vocabulary that never counts as a significant word by itself,
  regardless of length.

checkCompanyMatch, exact/near-exact role matches, and Chinese role-title
handling are unchanged.

Closes career-ops-hq#2671

…bit review on career-ops-hq#2672)

Addresses two CodeRabbit findings on PR career-ops-hq#2672:

1. Major — checkRoleMatchExact() only excluded single-word roles that were
   also generic stoplist words (isBareGenericRole), so a specific
   single-word role like "Engineer" still qualified as a standalone exact
   match against any message containing that word anywhere, recreating the
   uncorroborated false-positive path the PR was meant to close. Replaced
   isBareGenericRole with isSingleWordRole: any role that reduces to a
   single word — generic or not — now falls through to the partial-match
   path and requires company/domain corroboration in matchCandidates(),
   same as any other single significant word. Chinese compound titles
   (which carry no separators to split on) are explicitly exempted via a
   CJK check, preserving the existing Chinese-title exact-match behavior.

2. Minor — added a matchCandidates test exercising sender-domain
   corroboration on its own (no company-name text in the message), since
   the existing corroboration test only covered the company-name path.

Also added a regression test proving Chinese single-part titles still
match exactly, and one confirming checkRoleMatch's boolean convenience
path still reports a match for a single-word role — it's matchCandidates
that enforces corroboration, not checkRoleMatch itself.

…Rabbit review on career-ops-hq#2672)

normalizeStr(' ') === '', and String.prototype.includes('') is always true,
so a whitespace-only role title (e.g. "   ") produced a false "exact" match
against any text at all, bypassing the corroboration requirement entirely.
isSingleWordRole doesn't catch this case: splitting a whitespace-only string
on separators yields zero parts, not one.

Added an explicit rNorm-empty guard in checkRoleMatchExact and a regression
test confirming a whitespace-only role never matches arbitrary text.