Looking through the text of the R.gveda, it appears that words can begin with any basic consonant with the exception of some of the retroflex or aspirated stops: k, kh, g, gh, c, ch, j, t., n., t, d, dh, n, p, ph, b, bh, m, y, r, l, v, s", s., s, h. (Underlyingly, words do not begin with retroflex stops or n., and very few begin with s.; these appear on the surface courtesy of retroflection rules, which can apply across words within a phonological phrase.) More importantly, there are many clusters. It appears that in general an obstruent or h can be followed by a sonorant (I find kr, kl, kv, khy, gn, gm, gr, gl, ghn, ghr, chn, chm, chy, chr, chl, chv, jn~, jm, jy, jr, tm, ty, tr, try, tv, dy, dr, drv, dv, dhm, dhy, dhr, dhv, ny, nv, py, pr, pl, br, bhr, my, mr, ml, vr, vy, s"n, s"m, s"y, s"r, s"l, s"v, s.n, s.m, sn, sm, sy, sr, sv, hn, hr, hl, hv, hy). (The sequences with ch- are sandhi forms of underlying s"- after a dental stop or nasal.) A voiceless stop can be preceded by a sibilant, which must be homorganic in the case of coronals (s"c, s.k, s.t., s.t.h, s.ph, sk, st, str, sth, sp, sph). A voiceless unaspirated stop may also be followed by a sibilant (ks., ks.m, ks.v, ts, ps). The inventory of consonants that can follow a sonorant is more idiosyncratic: ny, nv, my, mr, ml, vr, vy. But clearly only sonorants may follow a sonorant, and only labial sonorants can be followed by a liquid.
There are some gaps in the generalized pattern. The sound l never follows a dental, which is a quite common restriction cross-linguistically (Clements 1990:316). The near absence of palatals and of voiceless aspirates in clusters can be referred to their fairly recent origin, where they developed from other sounds, primarily before vocoids. Dental stops are not found before n and labials are not found before v. Certainly homorganic clusters are universally discouraged in onsets, but these may just be accidental gaps, since Vedic does have homorganic jn~, sn, and jy.
The presence of word-initial clusters makes for a certain presumption against the CS theory. All things being equal, one would expect word-internal onsets and word-initial onsets to follow the same rules; strict CS is normally associated with a language lacking word-initial clusters. To harmonize these clusters with CS, one would have to assume a large set of word-initial appendices, even though word-initial appendices are not very common cross-linguistically, and these clusters look like typologically quite normal onset sequences. Or one might hypothesize a very strong preference for building codas, even though onsets are strongly favoured cross-linguistically.