fix: remediate 13 reachable security findings (cmdi, SSRF, PII, secret, error-leak, vuln-dep)

[Copilot]

Addresses a batch of 13 production-actionable security findings across the handler layer, spanning command injection, SSRF, PII leakage, hardcoded secrets, error disclosure, and a vulnerable dependency.

Command Injection (CWE-78) — cwe.go

• DiagnosticPing: replaced exec.Command("sh", "-c", "ping -c 1 "+host) with host validation + fixed-argv exec.Command("ping", "-c", "1", host)
• Both ping handlers now log command failures server-side and return a generic client message instead of raw command output

SSRF + Cleartext (CWE-918/319) — suspicious.go

• FetchTool: validates https-only scheme; replaces http.Get with a custom http.Client using ssrfSafeDial — a dialer that validates resolved IPs against private CIDR ranges at TCP connect time, eliminating the DNS TOCTOU/rebinding gap

var safeFetchClient = &http.Client{
    Transport: &http.Transport{DialContext: ssrfSafeDial},
}
// ssrfSafeDial resolves the hostname, checks each IP against privateNets,
// then dials — no window between check and connect.

LLM Prompt Injection (AI) — ai.go

• AIAnswer / AIAgentPlan: user-controlled content moved to a dedicated "role": "user" message instead of being concatenated into the system instruction string; raw prompt/tool-spec no longer echoed in the response

PII Leakage (DLP) — dlp.go

• SupportExport: SSN and DOB replaced with [REDACTED] in both the server log and the analytics POST payload

Hardcoded Secret — secrets.go

• Removed const syntheticGitHubToken = "ghp_reachtestbedsynthetic000000000000000000"
• CloudTokens reads GITHUB_TOKEN from the environment; returns "configured" or "" — never the token value itself

Error Disclosure (CWE-200) — ai.go, cve.go, suspicious.go

• SafeAIAnswer, ParseLanguage, and FetchTool error paths replaced err.Error() responses with generic client-safe messages; detail logged server-side

Vulnerable Dependency (CVE)

• golang.org/x/text bumped v0.3.7 → v0.3.8