The museum server's EmailAuthorizationResponse declares
passkeySessionID, accountsUrl, twoFactorSessionID, and
twoFactorSessionIDV2 without `omitempty`, so Go always sends them,
as "" when unset. The previous mock omitted the unset fields
entirely, which let the ?? -based dispatch pass in tests while the
real server's "" defeated it and dual-2FA logins fell through to
the unsupported passkey branch.
When an account has both passkeys and TOTP enrolled, the Ente server
returns passkeySessionID + twoFactorSessionIDV2 (deliberately not the
V1 twoFactorSessionID field, so old clients keep using passkeys).
quak only checked the V1 field, saw the passkey session, and threw
'Passkey authentication is not supported'. A CLI cannot do WebAuthn,
so it must take the TOTP path via the V2 session ID.
Adds fast-srp-hap (the same SRP library Ente's web client uses, pinned
to 2.0.4) as a runtime dependency.
Tests build a full mock Ente server using fast-srp-hap's SrpServer to
exercise real SRP-6a math end-to-end. The mock handles:
GET /users/srp/attributes
POST /users/srp/create-session
POST /users/srp/verify-session
POST /users/two-factor/verify
POST /users/ott
POST /users/verify-email
7 tests covering:
* SRP login completing successfully
* SRP login requiring TOTP (returns { kind: 'totp' })
* Wrong password (SRP M1 fails server-side checkM1)
* Email MFA fallback (returns { kind: 'emailOTP' })
* submitTOTP
* requestEmailOTP + submitEmailOTP
